WO2024248033A1 - Document management system and document management method using ai - Google Patents

Abstract

Provided is a mechanism for increasing business efficiency using summary generation by AI.　A document management system comprising: a document information acquisition means for acquiring first document information; a summary generation means for inputting the first document information acquired by the document information acquisition means to a trained machine learning model that has been trained using machine learning so as to generate a summary using document information as an input, and thereby acquiring a summary for the first document information from the machine learning model; and an output processing means for outputting the summary acquired by the summary generation means.

Description

Document management system and document management method using AI

[Related Applications]
This application claims priority to Japanese Patent Application No. 2023-088817, entitled “Document Management System and Document Management Method Using AI,” filed on May 30, 2023, the disclosure of which is incorporated herein by reference in its entirety.
The present disclosure relates to a document management system and a document management method using AI.

JP 2013-190991 A (Patent Document 1) is a background technology in this technical field. This publication states that "In a support method for streamlining response monitoring (listening to and evaluating speech) in call centers, the present invention aims to make it possible to streamline monitoring work even for a set of responses that deviate from the dialogue flow of the talk script. The spoken dialogue summarization device extracts multiple utterances from a set of dialogue texts, and extracts utterances that are mutually linked by implication relationships from the extracted multiple utterances as utterance clusters" (see abstract).

JP 2013-190991 A

The above-mentioned Patent Document 1 describes a mechanism of a voice dialogue summarization device that extracts multiple utterances from a collection of dialogue texts in a call center and extracts utterance clusters. However, this Patent Document does not consider a method of improving the efficiency of various types of customer service operations using telephone, chat, email, etc. by using AI-based summary generation.
Therefore, the present disclosure provides a mechanism for improving business efficiency by using AI-based summary generation.

In order to solve the above problems, for example, the configurations described in the claims are adopted.
The present application includes multiple means for solving the above-mentioned problems, and one example is characterized by having a document information acquisition means for acquiring first document information, a summary generation means for inputting the first document information acquired by the document information acquisition means into a trained machine learning model that has performed machine learning to generate a summary using document information as input, and acquiring a summary for the first document information from the machine learning model, and an output processing means for outputting the summary acquired by the summary generation means.

According to the present disclosure, a mechanism for improving business efficiency using summary generation by AI can be provided.
Problems, configurations and effects other than those described above will become apparent from the following description of the embodiments.

FIG. 1 is a diagram showing an example of the configuration of an entire customer response management system 1. As shown in FIG. FIG. 2 is a diagram showing an example of the hardware configuration of the management server 101. As shown in FIG. FIG. 3 is a diagram showing an example of the hardware configuration of the user terminal 102. As shown in FIG. FIG. 4 is a diagram showing an example of the hardware configuration of the administrator terminal 103. As shown in FIG. FIG. 5 is an example of a system configuration diagram of the summary generation module 214. As shown in FIG. FIG. 6 is an example of customer information 600 . FIG. 7 is an example of category information 700. FIG. 8 is an example of document information 800 . FIG. 9 is an example of summary information 900. FIG. 10 is an example of memo information 1000. FIG. 11 is an example of the response record information 1100. FIG. 12 is an example of FAQ information 1200. FIG. 13 is an example of auxiliary input information 1300. FIG. 14 is an example of a customer response process flow 1400 . FIG. 15 shows an example of a document information acquisition process flow 1500 . FIG. 16 is an example of a summary generation process flow 1600 . FIG. 17 is an example of a detailed summary generation process flow 1700 . FIG. 18 is an example of a process flow 1800 for generating a summary by category. FIG. 19 is an example of another process flow 1900 for generating summaries by category. FIG. 20 shows an example of a highlighting process flow 2000. FIG. 21 is an example of a customer information search screen 2100. FIG. 22 shows an example of a customer information inquiry screen 2200. FIG. 23 shows an example of a category selection input screen 2300. FIG. 24 is an example of a summary display screen 2400. FIG. 25 is an example of another summary display screen 2500. FIG. 26 is an example of a summary setting screen 2600. FIG. 27 shows an example of a reference information call setting screen 2700. FIG. 28 shows an example of a feedback reception screen 2800. FIG. 29 shows an example of a work status display screen 2900. FIG. 30 shows an example of a work status details display screen 3000. FIG. 31 is an example of another feedback acceptance screen 3100. FIG. 32 shows an example of a summary generation instruction screen 3200. FIG. 33 is an example of a summary display screen 3300. FIG. 34 shows an example of a customer response process flow 3400 using category information.

Hereinafter, the embodiments will be described with reference to the drawings.
This embodiment allows users, such as operators working at a call center, to easily perform customer service operations.
For example, we provide a system that allows a call center operator to easily execute a series of processes after a customer interaction by inputting information about the call content into the system and reflecting it in a CRM (Customer Relationship Management) system. This type of processing after a customer interaction is called ACW (After Call Work).
Examples of ACW that can be handled by the system include the process of entering and selecting various category information about the inquiry content, the process of creating a summary of the call content, and the process of storing the response status and method in the CRM system in association with customer information confirmed during the call.

By using this system, customer service operations can be centralized and ACW time can be dramatically reduced, resulting in more efficient call center operations.
This system is not limited to telephone response work at a call center, but can also be applied to customer response work using chat, e-mail, etc., and can also be applied to any document management work.

The customer response management system 1 comprises a plurality of user terminals 102 and a plurality of manager terminals 103, each of which is connected to a management server 101 via a network. Note that the network may be wired or wireless, and each terminal can send and receive information via the network.
The user terminal 102 is a terminal used by a user, such as an operator who handles customer service at a call center.
The administrator terminal 103 is a terminal that manages the multiple user terminals 102 and the work of many operators operating in the call center. The administrator terminal 103 may also have a function for managing the management server 101, such as changing the settings of the management server 101.

Each terminal and management server 101 of the customer response management system 1 may be, for example, a portable terminal (mobile terminal) such as a smartphone, tablet, mobile phone, or personal digital assistant (PDA), or a wearable terminal such as glasses/goggles, wristwatch, or clothing. They may also be stationary or portable computers, or servers located on the cloud or network. In terms of functionality, they may be a VR (Virtual Reality) terminal, an AR (Augmented Reality) terminal, or an MR (Mixed Reality) terminal. Or they may be a combination of multiple of these terminals. For example, a combination of one smartphone and one wearable terminal can logically function as one terminal. They may also be other information processing terminals.

Each terminal and management server 101 of the customer response management system 1 comprises a processor that executes an operating system, applications, programs, etc.; a main memory device such as a RAM (Random Access Memory); an auxiliary memory device such as an IC card, hard disk drive, SSD (Solid State Drive), flash memory, etc.; a communication control unit such as a network card, wireless communication module, or mobile communication module; input devices such as a touch panel, keyboard, mouse, voice input, and input based on motion detection from image capture by the camera unit; and an output device such as a monitor or display. The output device may be a device or terminal that transmits information to be output to an external monitor, display, printer, equipment, etc.

The main memory stores various programs and applications (modules), and the processor executes these programs and applications to realize each functional element of the overall system. Each of these modules may be implemented in hardware, for example by integration. Each module may be an independent program or application, or may be implemented as a subprogram or function within a single integrated program or application.

In this specification, each module is described as an entity (subject) that performs processing, but in reality, a processor that processes various programs, applications, etc. (modules) executes the processing.
Various databases (DB) are stored in the auxiliary storage device. A "database" is a functional element (storage unit) that stores a data set so that it can handle any data operation (e.g., extraction, addition, deletion, overwriting, etc.) from a processor or an external computer. The method of implementing the database is not limited, and may be, for example, a database management system, a file management system, spreadsheet software, or a text file such as XML or JSON.

FIG. 2 is a diagram showing an example of the hardware configuration of the management server 101. As shown in FIG.
The management server 101 is configured, for example, as a server deployed on a cloud.
However, the management server 101 may be constructed as an on-premise type in the same base as the user terminal 102 and/or the administrator terminal 103. Also, the management server 101 may be configured integrally with the user terminal 102 and/or the administrator terminal 103.
The main memory device 201 stores programs and applications such as a customer response processing module 211, a category selection module 212, a document information acquisition module 213, a summary generation module 214, a memo information acquisition module 215, a highlighting module 216, a feedback module 217, and a data search module 218, and the various functional elements of the management server 101 are realized by the processor 203 executing these programs and applications.

The customer response processing module 211 provides a function for managing the overall customer response processing performed by users (operators, etc.). In this embodiment, a telephone response in a call center will be described as an example, but the present invention is not limited to this and can also be applied to customer requests received via chat, email, inquiry forms, message transmission and reception services, etc.
Furthermore, the customer response processing module 211 can display the document information acquired by the document information acquisition module 213 and the summary generated by the summary generation module 214 together on one screen in a corresponding manner. In addition to the configuration in which two panels are displayed on one screen, it is also possible to display, for example, the document information and the summary by switching between them using tabs or the like. In this case, too, it can be said that the document information and the summary are displayed in a corresponding manner. Alternatively, it is also possible to simply display the contents of one panel.

The category selection module 212 identifies which category the customer inquiry belongs to. Examples of categories include the following:
Target business categories: finance, insurance, e-commerce, public energy (electricity, gas, water, etc.), tourism, bridal, etc.
Applicable product categories:
- For the bridal category, wedding venues, reception parties, honeymoons, etc.
- Insurance categories include property insurance, life insurance, education insurance, car insurance, fire insurance, etc.
Content category: application/procedure, questions (inquiries), consultations, requests, etc.
By designing the category information hierarchically, it becomes possible to flexibly respond to various types of customer service processing.
The category information is stored in the category information 700, but it may be registered in advance or may be freely written and registered by the user, etc. Also, the category selection module 212 may be configured to newly register, as category information, a character string extracted from terms used in the dialogue text.
The category information may be meta-information. Meta-information for a document or the like can be treated as category information.

The document information acquisition module 213 acquires the contents of a conversation with a customer as text data and stores it as document information. For example, the document information is generated by analyzing the voice of a telephone conversation with a customer and converting it into text.
In the case of communication with customers through chat, e-mail, inquiry forms, message sending and receiving services, etc., the contents of the text sent and received are stored as document information. In addition, it is not limited to the contents of the communication with customers, as long as any document information can be obtained.

The summary generation module 214 inputs document information and generates a summary of the document information. When using AI, the summary generation module 214 generates a summary for the input document information by inputting the document information into a trained machine learning model that has been trained to generate summaries using the document information as input. At this time, by inputting category information together with the document information, the accuracy of the generated summary can be improved and the summary content can be adjusted.
The memo information acquisition module 215 acquires and stores memos entered by the user while serving a customer. The memo may be a blank screen into which text is entered, or may have a predetermined format into which information is entered. Any text information may be entered. The memo may be acquired before or after serving a customer.

The highlighting module 216 makes it easier to handle customer inquiries by highlighting parts of document information, summaries, and memo information that satisfy certain conditions.
Highlighting may be achieved, for example, by making the portion of a document that meets a condition bold, underlining, changing the color, italicizing, shading, changing the font, blinking, surrounding the text with a line, changing the background color, or overlaying a color layer on top of the text. Highlighting may be achieved by making the text more noticeable or distinct than other portions of the text. The text to be highlighted may also be displayed on a separate screen or window.

The feedback module 217 receives user evaluations and corrections to the generated summary. The feedback module 217 may also receive corrections to document information and memo information. In this embodiment, these evaluations and corrections are collectively called feedback from the user.
The data search module 218 searches for the customer information 600, the FAQ information 1200, the input assistance information 1300, and the like during customer response operations.

The auxiliary storage device 202 includes various databases for storing various types of information.
Each database stores, for example, customer information 600, category information 700, document information 800, summary information 900, memo information 1000, and interaction record information 1100. Each piece of information will be described in detail later.

FIG. 3 is a diagram showing an example of the hardware configuration of the user terminal 102. As shown in FIG.
The user terminal 102 is composed of a terminal such as a smartphone, a tablet, a notebook PC, or a desktop PC.
The main memory device 301 stores programs and applications such as a management server cooperation module 310, and the processor 303 executes these programs and applications to realize each functional element of the user terminal 102.

The management server cooperation module 310 cooperates with the customer service processing module 211 of the management server 101 to execute the customer service processing provided by the management server 101. The management server cooperation module 310 displays various screens such as Figures 21 to 33 provided by the management server 101 on a display, which is one of the output devices 205.

The customer service processing module 211 of the management server 101 provides, for example, a customer service function from the cloud, and the management server cooperation module 310 displays screen information sent from the management server 101 on the display.
However, the configuration may be such that all of the functions of the programs and applications provided by the management server 101 are implemented on the user terminal 102, and the user terminal 102 realizes the customer response function alone, or the configuration may be such that some of the functions of the programs and applications are implemented on the user terminal 102, and the user terminal 102 and the management server 101 work together to realize the customer response function.

The auxiliary storage device 302 stores, for example, user terminal management information 320 .
The user terminal management information 320 stores, for example, information for identifying an operator who is a user using the user terminal 102 .
In addition, in the case where some of the functions of the management server 101 are implemented on the user terminal 102 side, the same information as the various information stored in the management server 101 may be stored in the user terminal management information 320 .

FIG. 4 is a diagram showing an example of the hardware configuration of the administrator terminal 103. As shown in FIG.
The administrator terminal 103 is configured as a terminal such as a smartphone, a tablet, a notebook PC, or a desktop PC.
The main memory device 401 stores programs and applications such as a user terminal management module 410 and a management server management module 411, and each functional element of the administrator terminal 103 is realized by the processor 403 executing these programs and applications.

The user terminal management module 410 collectively manages a plurality of user terminals 102 operating in the call center.
The user terminal management module 410 displays a screen that allows monitoring of the efficiency of ACW work, such as those shown in Fig. 29 or 30. A list of each call (interaction) can be confirmed and filtered, and the input of each interaction can be displayed. The information displayed on these screens is obtained by the user terminal management module 410 in cooperation with the customer response processing module 211 of the management server 101.

In addition, the user terminal management module 410 calculates and displays the distribution and statistics of the time spent on ACW for each important attribute, such as by time (month, week, etc.), by user, by category of customer service content, etc.
In addition, the user terminal management module 410 displays the ACW work log of each user, for example, as shown in FIG.

The management server management module 411 performs various settings for the management server 101. For example, the management server management module 411 can perform settings for summary generation using a summary setting screen 2600 in FIG.
It is also possible to configure the system so that all of the functions of the programs and applications provided by the management server 101 are implemented on the administrator terminal 103, and the administrator terminal 103 realizes the customer service function. It is also possible to configure the system so that some of the functions of the programs and applications are implemented on the administrator terminal 103, and the user terminal 102 and the administrator terminal 103 work together to realize the customer service function.

The auxiliary storage device 402 stores, for example, administrator terminal information 420 .
The manager terminal information 420 stores, for example, information for identifying the manager who uses the manager terminal 103, information for identifying the user terminal 102 and each user that is managed.
In addition, in the case where some of the functions of the management server 101 are implemented on the administrator terminal 103 side, the same information as the various information stored in the management server 101 may be stored in the administrator terminal information 420 .

FIG. 5 is an example of a system configuration diagram of the summary generation module 214. As shown in FIG.
There are various possible ways to implement the summary generation module 214, but FIG. 5 shows an example of a specific implementation using a machine learning module.
An API (Application Programming Interface) module 501 receives a request for generating a summary from the user terminal 102, acquires document information, and starts the summary generation process. The API module 501 also transmits the generated summary to the user terminal 102 via a Web server module 505. The API module 501 may be implemented as an independent API server.

The message queuing module 502 temporarily stores document information in a queue, and inputs the information to the ML module 503 in sequence.
The ML (Machine Learning) module acquires document information stored in the queue, generates a summary, and passes it to the queue. The ML module 503 may be implemented as an independent ML server.
The worker module 504 stores the summary received from the message queuing module 502 in the DB 511 in association with the document ID.
The Web server module 505 displays the summary results and the like as a UI (User Interface).
The storage device 512 stores multiple machine learning trained models, and by reading them into the ML module 503, the content of the generated summary can be changed.
A detailed summary generation process flow by the summary generation module 214 in FIG. 5 will be described later with reference to FIG.

6 to 13 show various types of information stored in the management server 101.
FIG. 6 is an example of customer information 600 .
The customer information 600 stores information about customers who call the call center.
The customer information 600 includes information to identify the customer, such as a user ID 601, a user display, a name, a telephone number, an email address, a date of birth, a gender, and an address, and values such as those shown as sample values 620 are entered for each item 610.
The user ID 601 is a unique ID, such as an automatically generated hash value, and is a primary key referenced by other information.
The user display (sometimes called a user title or user display name) is a user name that is displayed on the screen of the user terminal 102, the administrator terminal 103, or the like.

FIG. 7 is an example of category information 700.
The category information 700 is information selected according to the content of customer service, and stores information indicating the content of customer service.
The category information 700 includes a category ID 701, a subcategory ID 702, a subcategory ID 2 703, subcategory content 704, subcategory content 705, and subcategory content 2 706, and values such as those shown as sample values 720 are input for each item 710.
A category ID 701, a subcategory ID 702, and a subcategory ID2 703 are IDs that, either alone or in combination, uniquely identify the content of a customer response.

In the example of FIG. 7, Bridal001 of category ID 701 indicates wedding-related category content 704, Ceremony001 of subcategory ID 702 indicates wedding hall-related subcategory content 705, and Inquiry002 of subcategory ID 2 703 indicates inquiry-related subcategory content 2 706.
In this example, under the category "Wedding," for example, there are subcategories related to wedding venues, as well as subcategories related to reception parties, honeymoons, and the like.
In addition to inquiries, the wedding hall-related subcategories include an application sub-subcategory.

In another example, under the category of insurance, there are subcategories such as property insurance, life insurance, education insurance, automobile insurance, and fire insurance.
Furthermore, under the non-life insurance subcategory, there are sub-subcategories such as inquiries, applications, and insurance claims.
In this way, by designing the category information hierarchically, it becomes possible to flexibly respond to various types of customer service processing.

7, the categories have a three-level hierarchy, but they may have two levels, or four or more levels.Also, there may be no hierarchy, and all categories may be managed at only one level of category ID.
In addition, although the above description is based on the structure in which the hierarchy goes down in the order of weddings, wedding venue-related, and inquiries, this order may be reversed, for example, the hierarchy may go down in the order of inquiries, weddings, and wedding venue-related.

FIG. 8 is an example of document information 800 .
The document information 800 stores, for example, information about documents acquired by customer service.
The document information 800 includes a document ID 801, a document display 802, a recording date and time 803, and a document information storage URL 804, and values such as those shown as sample values 820 are input to each item 810.

The document ID 801 is identification information that uniquely identifies a stored document.
The document display 802 (sometimes called a document title 802 or a document display name 802 ) stores a description of the document, such as the title, displayed on the user terminal 102 or the administrator terminal 103 .
The recording date and time 803 indicates the date and time when the document was stored in the storage device. In the example of Fig. 8, "20220401141201" indicates "April 1, 2022, 14:12:01".
Furthermore, when a document is updated, the update date and time of the document and the last update date and time may also be stored.
The document information storage URL 804 indicates the location where the actual document is stored. Note that instead of a method of specifying a path to data in the form of a URL, a method of specifying a path to a file in storage inside the server may also be used.
Also, if the data size is not large, the information may be written directly.

In this embodiment, the document information stored in the document information storage URL 804 of the document information 800 may be referred to as "document information." For example, text data stored in the document information storage URL 804 may be referred to as "document information."
Furthermore, text data before being stored in the document information 800 may also be called "document information."

The text data stored here may be, for example, dialogue text, which is a text version of the content of a conversation between a customer and an operator during customer service. In addition, in the case of communication with a customer via chat, e-mail, an inquiry form, a message sending/receiving service, etc., the content of the text sent and received is stored as document information. In addition, any document information related to the content of the communication with the customer may be acquired and stored.
In addition, when the present embodiment is applied to a wide range of business operations other than call centers, it is sufficient to acquire and store some document information. Customer service is called by various names, such as support, inquiries, customer success, CX (Customer Experience), and contact centers.

FIG. 9 is an example of summary information 900.
The summary information 900 stores the summaries generated by the summary generation module 214 .
The summary information 900 includes a summary ID 901, a summary display 902, a document ID 903, a recording date and time 904, a last update date and time 905, and a summary information storage URL 906. Values such as those shown as sample values 920 are input to each item 910.

The summary ID 901 is identification information that uniquely identifies the stored summary.
The summary display 902 (sometimes called summary title 902 or summary display name 902 ) stores an explanation such as the title of the summary to be displayed on the user terminal 102 or the administrator terminal 103 .
The document ID 903 stores the value of the document ID 801 of the document information 800, and specifies which document information 800 the summary corresponds to.
Recorded date and time 904 indicates the date and time the summary was stored in the storage device.

The last update date and time 905 indicates the date and time when the summary was last updated after being modified. Note that in the case where the summary is updated multiple times, not only the last update date and time but also the update date and time of the summary may be stored.
Summary information storage URL 906 indicates the location where the actual summary is stored. Note that instead of specifying the path to the data in the form of a URL, a path to a file in storage within the server may be specified. If the data size is not large, the information may be written directly.
In this embodiment, when a summary identified by one summary ID 901 is updated multiple times, the update history of the summary or all versions of the summary can be stored in association with the same summary ID 901. Alternatively, a new summary ID may be assigned each time the summary is updated and stored.

FIG. 10 is an example of memo information 1000.
The memo information 1000 has a memo ID 1001, a memo display 1002, a document ID 1003, a recording date and time 1004, a last update date and time 1005, and a memo information storage URL 1006, and values such as those shown as sample values 1020 are input to each item 1010.
The memo ID 1001 is identification information that uniquely identifies the stored memo information.
The memo display 1002 (sometimes called a memo title 1002 or a memo display name 1002 ) stores an explanation such as a title of memo information displayed on the user terminal 102 or the administrator terminal 103 .
The document ID 1003 stores the value of the document ID 802 of the document information 800, and specifies which document information 800 the memo information corresponds to.
The recording date and time 1004 indicates the date and time when the memo information was stored in the storage device.

The last update date and time 1005 indicates the last update date and time after the memo information is corrected. Note that in addition to the last update date and time, the update date and time of the memo information when it is updated multiple times may also be stored.
The memo information storage URL 1006 indicates the location where the actual memo information is stored.
In this embodiment, when multiple updates are made to memo information identified by one memo ID 1001, the update history can be stored in association with the same memo ID 1001, and a new memo ID can also be reassigned each time an update is made.

FIG. 11 is an example of the response record information 1100.
The content of the response when the user responds to a customer is stored as a response record.
The response record information 1100 includes a response ID 1101, a response start date and time 1102, a response end date and time 1103, a person in charge 1104, a customer ID 1105, a category ID 1106, a summary ID 1107, an evaluation score 1108, and a customer input data storage URL 1109, and values such as those shown as sample values 1120 are entered for each item 1110.

The response ID 1101 is identification information that uniquely identifies a customer response.
The service start date/time 1102 and service end date/time 1103 indicate the date/time when the customer service started and ended.
The person in charge 1104 stores information for identifying the user (operator) who attended to the customer.
The customer ID stores information for identifying the customer served.

The category ID 1106 is information indicating the content of the customer service, and stores the category ID 701 of the category information 700 in Fig. 7. Note that the subcategory ID and subcategory ID2 below the category ID 1106 store information on the subcategory ID 702 and subcategory ID2 703 in Fig. 7, respectively.
The summary ID 1107 stores the information of the summary ID 901 in FIG. 9, and stores in association with what kind of summary has been generated based on the customer response.

The evaluation score 1108 stores the evaluation score given to the generated summary by the user or the administrator. The evaluation score 1108 may be stored in the summary information 900 instead of in the response record information 1100.
The customer input data storage URL 1109 stores information that identifies or explains the content of the customer response and the content of the processing input as a result. For example, in the case of a response to an application from a customer, information about the resulting application is written in the customer input data storage URL 1109.

FIG. 12 is an example of FAQ information 1200.
The FAQ information 1200 stores a list of frequently asked questions and their answers.
The FAQ information 1200 has an FAQ ID 1201 , an overview, a category ID 1202 , a subcategory ID, and an FAQ data storage URL 1203 , and values such as those shown as sample values 1220 are input for each item 1210 .

The FAQ ID 1201 is identification information that uniquely identifies the FAQ.
The summary stores information indicating the contents, such as the title of the FAQ.
The FAQ data storage URL 1203 indicates the location where the actual FAQ content is stored. In addition to the method of specifying the path to the data in the form of a URL, a method of specifying a path to a file in storage inside the server may also be used. Furthermore, if the data size is not large, the information may be written directly. For example, text information with questions and answers in separate columns may be stored as FAQ data.

The category ID 1202 is information indicating the content of the customer service, and stores the category ID 701 of the category information 700 in Fig. 7. Note that the subcategory ID and subcategory ID2 below the category ID 1202 store information on the subcategory ID 702 and subcategory ID2 703 in Fig. 7, respectively.
By identifying and storing a customer service category for the FAQ information 1200 in this manner, FAQ information related to the customer service category can be obtained and displayed on the user terminal 102 or the like.

FIG. 13 is an example of auxiliary input information 1300.
The auxiliary input information 1300 stores auxiliary input information for retrieving input contents to be input for each customer when serving the customer.
The auxiliary input information 1300 has an auxiliary input information ID 1301 , an overview, a category ID 1302 , a subcategory ID, and an input content storage URL 1303 , and values such as those shown as sample values 1320 are input for each item 1310 .
The auxiliary input information ID 1301 is identification information that uniquely identifies auxiliary input information such as a form to be input.
The summary stores information indicating the contents, such as the title of the input auxiliary information.

The category ID 1302 is information indicating the content of the customer service, and stores the category ID 701 of the category information 700 in Fig. 7. Note that the subcategory ID and subcategory ID2 below the category ID 1302 store information on the subcategory ID 702 and subcategory ID2 703 in Fig. 7, respectively.
In this way, by specifying and storing the customer service category for the auxiliary input information 1300, auxiliary input information related to the customer service category can be obtained and displayed on the user terminal 102 or the like.

FIG. 14 is an example of a customer response process flow 1400 .
The customer service processing module 211 starts customer service (step 1410). Specifically, when an operator (user) receives a call and starts customer service, the operator may be notified of the start of customer service, or the operator may input the start of customer service.
The document information acquisition module 213 converts the contents of the call into text during the call and acquires the document information (step 1420). The converted text of the call is sometimes called a dialogue text. The document information acquisition process will be described later.

Next, the customer interaction processing module 211 identifies the customer on the other end of the call (step 1430).
Specifically, the customer response processing module 211 calls the data search module 218, and the data search module 218 displays the contents of the customer search field 2140 on the customer information search screen 2100 of FIG.
The operator searches for customer information while looking at the contents of the call and the dialogue text, and if a corresponding customer is found, the operator specifies that customer as the call partner. If a corresponding customer is not found and a new customer is registered, this will be described later.

FIG. 21 is an example of a customer information search screen 2100.
In the dialogue text field 2110, the contents of the conversation that have been converted into text in the document information acquisition process 1420 are displayed in almost real time.
The memo field 2120 is a section where the operator inputs memos when responding to a customer. The memo information entered and acquired here may be called second document information.

The customer search field 2140 is a display area where the operator can input keywords for searching for a customer and search for customer information registered in the customer information 600 .
The customer response processing module 211 acquires information on the telephone number of the customer who is responding and displays it in the incoming call information 2141. Note that the configuration may be such that the operator inputs the telephone number displayed on the telephone into the incoming call information 2141.

Methods for identifying customer information include, for example:
The customer response processing module 211 automatically inputs customer information from the telephone number.
- The operator enters customer information by searching and selecting.
- Manually entered by the operator.
It is possible that this is the case.

On the other hand, the customer response processing module 211 may be configured to assist with input based on the contents of the dialogue text. For example, the customer response processing module 211 may input, into the name field of the search conditions 2142, name information 2111 that is likely to be a customer among information 2111, 2112 relating to personal names in the dialogue text displayed in the dialogue text field 2110. For example, the name information 2111 that appears first among the words uttered by the customer is input into the name field. Note that the configuration may also be such that the customer's name heard by the operator during the dialogue is input into the name field of the search conditions 2142.
The customer response processing module 211 can also extract named entities from the dialogue text and infer and input customer information from them. For example, customer information can be inferred and input from the name, phone number, content, etc. displayed in the dialogue text using a rule-based technique using a dictionary or the like, or a machine learning-based technique.

In addition, as the search conditions 2142, input of information about the customer, such as name, date of birth, telephone number, and postal code, can also be accepted.
The highlighting module 216 of the management server 101 highlights parts of the dialogue text displayed in the dialogue text field 2110 that relate to attributes that are likely to be important for a customer search, such as name, date, time, phone number, location, contract number, etc. If there is an item in the highlighted content that corresponds to the search condition 2142, the highlighting module 216 inputs the name, phone number, etc. that corresponds to the relevant part to complete it.
Conversely, when the operator enters information such as a name and telephone number heard over the telephone into the search conditions 2142, the highlighting module 216 can highlight the corresponding portion in the dialogue text.

When the data search module 218 detects that the operator has selected the search button 2143, it searches the customer information 600 using the incoming call information 2141 and the information entered in the search conditions 2142 to check whether a corresponding record exists. If a corresponding record is found in the customer information 600, the data search module 218 displays the customer information inquiry screen 2200 in FIG. 22.

The customer information display field 2150 displays customer information registered in the customer information 600, and when customer information is found by a search, the customer information found in the customer information display field 2150 may be displayed and highlighted.
If no corresponding customer information is found in the customer information 600, the customer response processing module 211 accepts the selection of a new customer registration button 2160, and displays a registration screen for new customer information, taking over the information entered in the incoming call information 2141 and the search condition field 2142. The new customer registration may also be provided with a function for completing the information from the dialogue text, and a function for highlighting the completed portion in the dialogue text.
In this embodiment, the data search module 218 displays information acquired from the customer information 600 in the customer search field 2140 , but this display may be entirely performed by the customer response processing module 211 .

FIG. 22 shows an example of a customer information inquiry screen 2200.
The data search module 218 displays customer information 2241 found by the search in a customer introduction section 2240 .
The highlighting module 216 highlights information corresponding to the customer information displayed in the displayed customer information 2241 on the dialogue text in the dialogue text field 2210 .
With this configuration, it is possible to easily confirm where in the dialogue text the customer information found by the search or the customer information to be newly registered is described.

The customer response processing module 211 stores a customer ID corresponding to a call partner customer identified by search or new registration. It is also possible to configure the data search module 218 to search the customer information 600 based on the incoming telephone number or the name extracted from the dialogue text to identify the customer without the user performing a search.
Also, the data search module 218 may be configured to display candidate customer names, etc., and the customer may be specified by the user inputting a confirmation.

FIG. 32 shows an example of a summary generation instruction screen 3200.
When the customer response processing module 211 detects that the summary start button 3235 has been pressed, it transmits the dialogue text, which is document information, to the summary generation module 214, and the summary generation module 214 executes the summary generation process (step 1440).
A specific summary generation process will be described in detail with reference to FIG.

33 is an example of a summary display screen 3300. The summary generated by the summary generation module 214 is displayed.
In the example of the displayed summary 3330, the customer name 3332 and the name of the person who answered the phone 3331 are highlighted, and important matters in the generated summary 3333 are also highlighted. Figures 32 and 33 show examples in which the input document information and the summary generated from it are each displayed on one panel, but these may be displayed on two panels in one window. Also, Figures 32 and 33, each showing only one panel, may be displayed side by side or vertically in one window.

The customer response processing module 211 associates the summary generated by the summary generating module 214 with the content of the customer response and stores them in the response record information 1100 (step 1450).
Specifically, the customer response processing module 211 generates a response ID 1101 for each customer response, and stores in the response record information 1100 the staff ID of the staff who responded, the customer ID indicating the customer identified in step 1430, the summary ID indicating the summary generated in step 1440, the response start date and time, and the response end date and time in association with each other.

If the feedback module 217 receives a rating for the summary, the rating information may also be stored.
Furthermore, when an application or the like occurs as a result of customer interaction, the input data for the procedure of the application or the like can be reflected in the CRM system, etc. In such a case, information specifying or explaining the content of the customer interaction and the content of the processing input as a result can be stored in the customer input data storage URL 1109 of the interaction record information 1100.
When an operator modifies the summary or updates the memo, the history of the changes may be stored in the interaction record information 1100, the summary information 900, the memo information 1000, or the like.

FIG. 34 shows an example of a customer response process flow 3400 using category information.
A step 3440 for accepting a selection of category information is added to the customer response process flow 1400 in FIG.
The processing of steps 3410 to 3430 by the customer response processing module 211 is similar to steps 1410 to 1430 in FIG. 14, and therefore a description thereof will be omitted.

Next, the customer response processing module 211 displays a category selection input screen 2300 shown in FIG. 23, and receives a selection of a category for the response content from the operator (step 3440).
Categories are classifications of customer service contents, and are stored in category information 700 shown in Fig. 7. In response to the selected category, the customer service processing module 211 acquires and stores information for identifying the category, such as the corresponding category ID, subcategory ID, and subcategory ID 2. As described above, the category information can be freely written by the user, or meta information can be registered, added, or used as category information.

FIG. 23 shows an example of a category selection input screen 2300.
The inquiry content input field 2350 displays fields for inputting or selecting categories such as case classification, content, and response method.
The caller classification is information for identifying what kind of person the customer is, such as an applicant, spouse, father, mother, contract holder, insured person, prospective customer, other party involved in an accident, dealer, financial institution, etc. The caller classification may be treated as a category.
Specifically, the customer response processing module 211 calls the category selection module 212, and the category selection module 212 displays the contents of the inquiry content input 2350 on the category selection input screen 2300 of FIG.

In the example of FIG. 23, the category "bridal", the subcategory "ceremony hall", and the sub-subcategory "inquiry" are selected according to the content of the customer service. When the set button 2352 is pressed after the category is selected, the customer service processing module 211 acquires and stores the category ID 701, subcategory ID 702, and subcategory ID2 703 shown in the example of FIG. 7 as the IDs corresponding to these categories.

The "content" and "type of response method" in the inquiry content input 2350 in FIG. 23 are displayed differently depending on the selected "case category." For example, if "insurance" is selected as the case category, "automobile" and "fire" will be displayed, indicating automobile insurance and fire insurance, which are subcategories of this insurance.

The highlighting module 216 can highlight character strings such as keywords and sentences related to the category in the dialogue text in the dialogue text field 2310 .
For example, in the example of Fig. 23, keywords related to bridal, such as "bridal,""ceremony in Hawaii," and "bride and groom," are highlighted. By checking the highlighted character strings, the operator can easily determine which category the content of the customer service belongs to, which has the effect of facilitating category selection.

Furthermore, keywords and sentences related to the categories registered in the category information 700 can be registered in advance, or related keywords and sentences can be acquired each time.
The category selection module 212 analyzes the keywords and sentences in the dialogue text, determines which category registered in advance in the category information 700 they relate to, and automatically selects the input in the inquiry content input field 2350 according to the keywords and sentences displayed in the dialogue text.
It is also possible to input dialogue text information and classify the entire sentence into categories on a rule-based basis using dictionary information or the like, or using a machine learning model.

Furthermore, by linking the category selection module 212 and the highlighting module 216, for example, when the operator selects "application" as the response method type, keywords related to "application" are highlighted in the dialogue text, and when the operator switches the dialogue method type to "inquiry," keywords related to "inquiry" are highlighted in the dialogue text.
Similarly, if the case category is changed from bridal to insurance, insurance-related keywords will be highlighted.

The category selection module 212 and the highlighting module 216 cooperate with each other to enable various implementations, such as the following:
The category selection module 212 can extract character strings such as keywords and sentences related to categories, which are classifications of the contents of customer service, from the dialogue text, and automatically select a category based on the extracted character strings. It is also possible to input dialogue text information and classify the entire text into categories on a rule-based basis using dictionary information or the like, or using a machine learning model.

In addition, the highlighting module 216 highlights character strings such as keywords and sentences related to a category from within the dialogue text, and the category selection module 212 can automatically select a related category based on the highlighted character strings.
In addition, the highlighting module 216 can highlight character strings related to the category contents in response to the selection of the category contents. When the selection of the category contents is changed, the highlighting module 216 can highlight character strings related to the changed category contents.
It is also possible to input dialogue text information and select a category from the entire text. It is also possible to input dialogue text information and select a category from the entire text on a rule-based basis using dictionary information or the like, or using a machine learning model.

In this embodiment, the category selection module 212 acquires the category information stored in the category information 700 and displays the content in the inquiry content input field 2350 based on this information, but this display may be entirely performed by the customer response processing module 211.

When the customer response processing module 211 detects that the summary start button 2335 has been pressed after the category has been selected, it transmits the dialogue text, which is document information, to the summary generation module 214, and the summary generation module 214 executes the summary generation process (step 3450).
A specific summary generation process will be described in detail with reference to FIG.
It should be noted that the selection of a category is not essential, and the summary generation process may be executed even if no category is selected. Also, a category may be automatically selected in the background by the category selection module 212 based on the contents of the document information, without displaying the category selection input screen 2300 as shown in FIG.

The customer response processing module 211 associates the summary generated by the summary generating module 214 with the content of the customer response and stores them in the response record information 1100 (step 3460).
Specifically, the customer response processing module 211 generates a response ID 1101 for each customer response, and stores in the response record information 1100 the staff member ID of the staff member who responded, the customer ID indicating the customer identified in step 3430, the category ID indicating the category selected in step 3440, etc., the summary ID indicating the summary generated in step 3450, the response start date and time, and the response end date and time in association with each other.

If the feedback module 217 receives a rating for the summary, the rating information may also be stored.
Furthermore, when an application or the like occurs as a result of customer interaction, the input data for the procedure of the application or the like can be reflected in the CRM system, etc. In such a case, information specifying or explaining the content of the customer interaction and the content of the processing input as a result can be stored in the customer input data storage URL 1109 of the interaction record information 1100.
When an operator modifies the summary or updates the memo, the history of the changes may be stored in the interaction record information 1100, the summary information 900, the memo information 1000, or the like.

FIG. 15 shows an example of a document information acquisition process flow 1500 .
Although telephone customer service is assumed, the present invention is not limited to this, and the document information acquisition process can be applied to anything that can acquire document information.
The document information acquisition module 213 receives input of telephone voice during telephone answering (step 1510).
The document information acquisition module 213 analyzes the voice information and converts it into text information (step 1520). Note that the content of the conversation with the customer converted into text is called a conversation text.
The document information acquisition module 213 displays the text information as dialogue text in, for example, the dialogue text field 2110 in FIG. 21 (step 1530).

If there is no need to store the text information, it is discarded (No in step 1540); if there is a need to store it (Yes in step 1540), the document information acquisition module 213 generates a document ID, associates it with the text information, and stores it in the document information 800 (step 1550).

The text information stored in document information 800 is not limited to dialogue text, but the contents of text sent and received in response to interactions with customers via chat, email, inquiry forms, message sending and receiving services, etc. are stored as text information. Any other document information related to the content of interactions with customers may be acquired and stored. Document information may also be acquired by linking with some other service, by being copied and brought over after a phone call, etc. Document information may also be acquired by linking with some other service, by being copied and brought over after a phone call, etc.
Furthermore, when this embodiment is widely applied to business other than call centers, it is sufficient to acquire and store some document information.

FIG. 16 is an example of a summary generation process flow 1600 .
The summary generation module 214 obtains document information (step 1610).
Specifically, the summary generation module 214 obtains text information (called document information) from a location referenced by the document information storage URL 804 of the document information 800 .
Alternatively, the dialogue text acquired by the document information acquisition module 213 may be acquired, or any other document information may be acquired.

The summary generation module 214 inputs the acquired document information (text information) into the trained machine learning model and generates a summary (step 1620).
The summary generation module 214 stores the generated summary in the summary information 900 in association with the generated document ID (step 1630).
Although the input for generating a summary is assumed to be the dialogue text or memos, other inputs such as FAQs referenced during a call or logs of other services used may also be referenced.

FIG. 17 is an example of a detailed summary generation process flow 1700 .
There are various possible ways to implement the summary generation module 214, but FIG. 5 shows an example of a specific implementation using a machine learning module, and the summary generation process using this configuration will be described.
The API module 501 obtains the text information (step 1705).
The API module 501 generates a document ID and stores the document ID and the text information in the document information 800 (step 1710).
The API module sends a message including the document ID and the text information to the message queuing module 502 (step 1715).

The ML module 503 inputs the message obtained from the message queuing module 502 into the machine learning model obtained from the storage device 512 to generate a summary (step 1720).
The ML module 503 sends a pair of the document ID and the summary result to the message queuing module 502 (step 1725).
The worker module 504 stores the document ID and the summary result in the summary information 900 (step 1735).

The API module 501 acquires the summary results (called summary information) stored in the summary information 900 (step 1740).
The Web server module 505 obtains the summary information from the API module 501 (step 1745).
The web server module 505 outputs the summary information (step 1750).
As the machine learning model, a large-scale language model can be used, which is a highly accurate generative model, but is not limited to this. For example, a machine learning model that generates summaries by character string extraction may be used.

FIG. 18 is an example of a process flow 1800 for generating a summary by category.
In the category-specific summary generation process flow 1800, when generating a summary, by further adding category information as an input, a summary suitable for the category can be generated.
The summary generation module 214 obtains document information (step 1810).
The summary generation module 214 obtains 1820 the category information obtained by the category selection module 212 .
The summary generation module 214 inputs the acquired document information (text information) and category information into the trained machine learning model, thereby generating a summary related to the category information for the input document information (step 1830).
The summary generation module 214 stores the generated summary in the summary information 900 in association with the generated document ID (step 1840).

The trained machine learning model can be a trained machine learning model that has been subjected to machine learning using category information indicating the category to which the document information belongs and multiple pieces of document information belonging to this category as training data.
In the example of selecting category information in step 3440 of the customer interaction processing flow 3400 using category information in Fig. 34, a category related to the dialogue text has already been selected. Therefore, by inputting document information of the dialogue text together with category information to this machine learning model, the summary generation module 214 can generate a summary optimized for this document information.
This processing flow uses a machine learning model that is trained by machine learning a set of category information and document information, and a single machine learning model can output a summary that corresponds to the input category.

As described above, the advance selection of a category is not essential, and the summary generation process may be executed even if no category is selected. Also, a category may be automatically selected in the background by the category selection module 212 based on the contents of the document information, without displaying the category selection input screen 2300 as shown in FIG.
Alternatively, instead of using category information itself, document information may be input into a machine learning model to determine the document category and then generate a summary corresponding to that category.

FIG. 19 is an example of another process flow 1900 for generating summaries by category.
In another category-specific summary generation process flow 1900, the machine learning model to be used is switched depending on the category information, thereby making it possible to generate summaries appropriate for the category.
The summary generation module 214 obtains document information (step 1910).
The summary generation module 214 obtains the category information obtained by the category selection module 212 (step 1920).

The summary generation module 214 selects a machine learning model that corresponds to the category information (step 1930). For example, if a separate server is prepared for each machine learning model, the document information is sent to the machine learning server that corresponds to the category information.

The summary generation module 214 inputs the acquired document information (text information) into a machine learning model selected corresponding to the category information, thereby generating a summary for the input document information (step 1940).
The summary generation module 214 stores the generated summary in the summary information 900 in association with the generated document ID (step 1950).

The trained machine learning model can be a trained machine learning model that has undergone machine learning using multiple document information belonging to a specific category as training data, and a trained machine learning model can be prepared for each category.
These trained machine learning models are stored in the storage device 512 in FIG. 5, and in response to the selection of a machine learning model, the selected machine learning model is read from the storage device 512 to the ML module 503.
In addition, a configuration in which an ML module 503 is prepared for each machine learning model optimized for a category and the ML module 503 is switched may also be used.

As described in the selection of category information in step 3440 of the customer interaction processing flow 3400 in Fig. 34, a category related to the dialogue text has already been selected. Therefore, by inputting the document information of the dialogue text to a machine learning model optimized for each category selected based on the category information, the summary generation module 214 can generate a summary optimized for this document information.
In this processing flow, a summary corresponding to the input category can be output by switching the machine learning model trained for each category based on the category information.

It should be noted that the selection of a category in step 3440 of the customer response processing flow 3400 in Fig. 34 is not essential, and the summary generation processing may be executed even if no category is selected. Also, a configuration may be adopted in which a category is automatically selected in the background by the category selection module 212 based on the contents of the document information, without displaying the category selection input screen 2300 as shown in Fig. 23.
That is, according to this embodiment, it is possible to provide a mechanism for explicitly switching summaries depending on the category information of the contents of a call.

FIG. 20 shows an example of a highlighting process flow 2000.
The highlighting module 216 obtains text information (called document information) from the document information 800 (step 2010).
The highlighting module 216 obtains a summary (referred to as summary information) from the summary information 900 (step 2020).
The highlighting module 216 obtains the written memo information from the memo information 1000 (step 2030).

The highlighting module 216 performs analysis, for example, by natural language processing, on each piece of acquired information, and extracts one or more characteristic character strings (step 2040).
The highlighting module 216 highlights the characteristic character string that satisfies various conditions described below, or a character string that includes this characteristic character string (step 2050).
A characteristic character string is a word or a character string containing a word in text data, and does not include, in principle, a character string containing only a particle. However, it is not limited to this, and may be a combination of words containing a particle, a character string containing a particle, or any other character string or sentence. A keyword is also one type of characteristic character string.
However, for example, it is also possible to highlight only particles in parts where there is a possibility of a particle error or where it is desired to emphasize a grammatical error.

Note that highlighting may be achieved, for example, by highlighting the portion of a document that meets a condition in bold, underlining, changing the color, italics, shading, changing the font, blinking, surrounding the text with a line, changing the background color, overlaying a color layer on the text, etc. Highlighting may be achieved by making the text stand out or distinguishable from other portions of the text.
In addition to highlighting characters in the text in some way, you can also highlight them by displaying them in a separate frame. For example, you can display a space on a separate screen or window that contains only the date information you want to highlight.

In the above description, the highlighting module 216 acquires information from the document information 800, summary information 900, and memo information 1000 stored in the storage device and performs highlighting, but this is not limiting. For example, the document information acquisition module 213, the summary generation module 214, and the memo information acquisition module 215 may acquire information from the document information 800, the summary information 900, and the memo information 1000, respectively, and the highlighting module may perform the highlighting process.
Also, the customer response processing module 211 may acquire various information, and the highlighting module may perform the highlighting process.

Furthermore, the highlighting module 216 may be configured to highlight the text information (dialogue text) acquired by the document information acquisition module 213, the summary information generated by the summary generation module 214, and the memo information acquired by the memo information acquisition module 215, instead of the document information 800, summary information 900, and memo information 1000 already stored in the storage device.
Further, the customer response processing module 211 may be configured to perform highlighting in cooperation with the highlighting module 216, or the customer response processing module 211 itself may have the functionality of the highlighting module 216.

FIG. 24 is an example of a summary display screen 2400.
After selecting a category in FIG. 23, a summary is generated by pressing summary start button 2335 and the summary is displayed in summary text field 2430.
The summary displayed here was generated by inputting the document information displayed in the dialogue text field 2410 into a machine learning model that generates a summary corresponding to the category selection.

The customer response processing module 211 displays three types of information - the first document information (dialogue text) acquired by the document information acquisition module 213, the summary acquired by the summary generation module 214, and the second document information (memo information) acquired by the memo information acquisition module 215 - in the dialogue text field 2410, summary field 2430, and memo field 2420, respectively, and outputs them in correspondence with each other using a three-panel display. The description in the dialogue text 2410 is a text version of the audio information of a telephone conversation, so there are some parts that have not been accurately transcribed. However, even when dialogue text containing such errors is used as input, the summary generation module 214 is able to generate a correct summary 2430.

It should be noted that all three panels are not necessary, and it is also possible to configure the system so that no dialogue, no summary, or no memo column is displayed, in which case any two of the dialogue text column 2410, summary column 2430, and memo column 2420 are displayed in association with each other. Also, it is possible to display each of the dialogue, summary, and memo columns individually.
Furthermore, it is not necessarily required to use a three-panel or two-panel display. For example, even if three types of information or any two types of information are displayed by switching between them using tabs or the like, it can be said that the configuration displays each piece of information in a corresponding manner.
Note that memo information may be used as the first document information, and dialogue text, chat, e-mail, an inquiry form, a message transmission/reception service, or the like may be used as the second document information.

The highlighting module 216 highlights the common name of the person in the dialogue text and the summary sentence, “Patent Jiro,” by underlining in 2411 and 2431 .
Even though they are not in common at all, "Eliza Hanako" 2432 in the summary and "Eliza Hanako" 2422 in the memo section are highlighted with an underline in comparison with the corresponding person's name "Eliza Hanako" 2412 in the dialogue text.
In addition, in the dialogue text, the statement 2413 "Airfare, hotel fees, etc. for other persons are not included" and the corresponding statement 2433 "Hotel fees, etc. for other persons are not included" are both highlighted in color.
Additionally, the date February 2, 2022 appears in both the dialogue text and the note, and is highlighted with an underline in each.

In this example, keywords in the dialogue text that are common to the summary are highlighted. In other words, the highlighting module 216 performs highlighting when a characteristic string corresponding to the first document information (dialogue text) and the summary exists as a condition. The highlighting module 216 highlights the characteristic string that corresponds to the summary from among one or more characteristic strings included in the first document information (dialogue text).
Additionally, it is also possible to highlight characteristic character strings that are commonly present in the dialogue text, summary text, or notes.

It is also possible to highlight corresponding keywords between the summary and the memo. In other words, the highlighting module 216 performs highlighting when a characteristic string corresponding to the second document information (memo information) and the summary is present as a condition. The highlighting module 216 highlights a characteristic string that corresponds to the second document information (memo information) and the summary among one or more characteristic strings included in at least one of the first document information (dialogue text), the second document information (memo information), and the summary.
Also, the dialogue text and the summary may be expressed differently, but the corresponding parts can be highlighted.

Pressing the setting button 2436 transitions to the summary generation setting screen of Fig. 26. Here, the settings can be changed, such as changing the category selection, and by pressing the summary start button 2435 again, a summary based on the changed settings is regenerated.
The summary generation setting screen may be configured to be displayed as a separate screen as shown in FIG. 26, or may be displayed on the summary output screen as shown in FIGS.

FIG. 25 is an example of another summary display screen 2500.
The parts that differ from FIG. 24 are highlighted.
In this example, statement 2511 in dialogue text 2310, "Airfare for the bride and groom is included, but airfare and hotel costs for others are not included," and the corresponding statement 2531 in summary sentence 2530, "Airfare for the bride and groom is included, but hotel costs for others are not included," are highlighted with underlines.

Additionally, keywords that are present in the notes but not in the summary can be highlighted in the dialogue information or notes.
That is, the highlighting module 216 highlights characteristic character strings that are present in the second document information (memo information) but not in the summary, among one or more characteristic character strings included in at least one of the first document information (dialogue text) and the second document information (memo information), that are present in the second document information (memo information) but not in the summary.

In the example of FIG. 25, the statement 2522 "I would like to add two more people" written in the memo field 2520 is present in the memo but is not present in the summary text 2530. Therefore, the highlighting module 216 highlights the corresponding statements 2512 and 2522 "I would like to add two more people" in the memo and dialogue text.

It can also highlight keywords in the summary that are present in the summary but not in the dialogue information or notes.
That is, the highlighting is performed when a characteristic character string is present in the summary but not in the first document information (dialogue text) or the second document information (memo information) under the condition that this condition is met. The highlighting module 216 highlights, of one or more characteristic character strings included in the summary, characteristic character strings that are present in the summary but not in the first document information (dialogue text) or the second document information (memo information).
In the example of Fig. 25, a statement 2538 "decided to change the wedding plan" exists in the summary but does not exist in the dialogue text or the memo. Depending on the machine learning model, a sentence that does not exist in the input dialogue text may be generated, but since a sentence containing such a character string may be different from the facts, highlighting the sentence containing such a character string gives the user an opportunity to correct it.

25, a statement 2533 of "future policy direction" written in the summary section 2530 exists in the summary, but does not exist in the memo section 2520. Therefore, the highlighting module 216 highlights the statement 2513 of "future policy direction" corresponding to this statement in the dialogue text.
It can also highlight strings of characters that are present in the summary and notes but not in the dialogue text.
It can also highlight strings of characters that are present in the summary and notes but not in the dialogue text.
It is also possible to highlight areas where the AI is unsure.

In other words, the condition is that among the summaries generated by the summary generation module 214, there is a possibility that the accuracy of the summaries is not high, and if this condition is met, the summaries are highlighted. For example, the judgment can be made according to the probability distribution of the model that outputs the summaries. For example, the summaries are highlighted when the probability distribution is constant and there is no significant difference in vocabulary selection. It is also possible to make the judgment by creating a separate learning model for discrimination.
The highlighting module 216 highlights, among the first document information (dialogue text), the second document information (memo information), or one or more characteristic strings contained in the summary generated by the summary generation module 214, those which may be of low accuracy.

Furthermore, the summary selection tag 2537 is a tab for comparing and displaying multiple summary sentences output by the machine learning model. The display method may be to display at least two or more summary sentences in parallel, or to switch by selecting a button or to switch in a pull-down format. The display method may be to display at least two or more summary sentences in parallel, or to switch by selecting a button or to switch in a pull-down format.
In addition, summaries generated by multiple machine learning models may be displayed for each tab, or summaries generated according to different categories may be displayed.

The other highlighting module 216 provides a mechanism for making it easier for people to review information by referring to input and output information through highlighting.
The following are mechanisms that make it easier for people to review input and output individually.
-Highlight important points (dates, names, contract numbers, etc.).
・Highlight areas where the AI is unsure.
- Provide a mechanism for easy reference and review between input and output.
- Highlight parts that are in the summary but have no similar input (dialogue text, notes, etc.).
Whether or not the points are close to each other can be determined by the following method.
- Judgment based on rules - Judgment based on the degree of match or similarity in some unit such as strings or words - Judgment based on dictionaries, etc. - Judgment based on machine learning technology - Similarity of distributed representations (vector representations) - Highlights which parts of the Input were used in judgment and summarization using a machine learning model.
-Highlight what is in your notes but not in the summary.
- Highlight parts of the dialogue that are likely to be included in the summary. Whether or not something is likely to be included can be determined using the following method.
- Generate multiple summaries and determine the strings in those summaries as candidates. - Use a machine learning model/rule to determine the importance separately.

After review, features for suggesting modifications include:
- Suggest corrections by using a thesaurus or machine learning to calculate the similarity between words and phrases in units of words or phrases. For example, the word "meeting" is highlighted in the summary, and "conference", "meeting", "MTG", and "meeting" are displayed as correction candidates that can be selected. Possible display formats include popping up correction candidates, displaying them in a separate tab or screen, or superimposing correction candidates when the cursor such as the mouse is moved over the highlighted portion. In addition, a configuration may be adopted in which correction suggestions are displayed for parts where the AI is not confident. In addition, a configuration may be adopted in which the user specifies an arbitrary location by clicking, etc., and correction candidates are displayed.
- For commonly confused words, phrases, etc., conversion rules are created using past data or manually, and then suggested.
- Propose other candidates that the AI could have output in various units such as words, phrases, and sentences. For example, propose the one with the next highest probability of generation. For example, among the candidates generated by the machine learning model, display other ones, such as the one with the next highest probability, as correction suggestions. The correction points may be, for example, points where the AI is not confident, or may be configured so that the user specifies any location by clicking, etc.
For example, in the summary example of Fig. 24, for the expression "A quote for Hawaii was provided," a candidate such as "A quote for Hawaii was mailed the other day" is presented. This correction suggestion may have a function for accepting feedback if the suggestion is incorrect or not very good. For example, a "check mark" indicating that the suggestion is accepted or a "cross mark" indicating that the suggestion is not accepted may be displayed below the correction suggestion, and a selection from the user may be accepted.
Alternatively, instead of accepting or rejecting a correction suggestion, a "good" or "bad" mark may be displayed to receive an evaluation of whether the suggestion is good or bad, and evaluation from the user may be received.

After review, features for suggesting modifications include:
- Suggest corrections by using a thesaurus or machine learning to calculate the similarity between words and phrases in units of words or phrases. For example, the word "meeting" is highlighted in the summary, and "conference", "meeting", "MTG", and "meeting" are displayed as correction candidates that can be selected. Possible display formats include popping up correction candidates, displaying them in a separate tab or screen, or superimposing correction candidates when the cursor such as the mouse is moved over the highlighted portion. In addition, a configuration may be adopted in which correction suggestions are displayed for parts where the AI is not confident. In addition, a configuration may be adopted in which the user specifies an arbitrary location by clicking, etc., and correction candidates are displayed.
- For commonly confused words, phrases, etc., conversion rules are created using past data or manually, and then suggested.
- Propose other candidates that the AI could have output in various units such as words, phrases, and sentences. For example, propose the one with the next highest probability of generation. For example, among the candidates generated by the machine learning model, display other ones, such as the one with the next highest probability, as correction suggestions. The correction points may be, for example, points where the AI is not confident, or may be configured so that the user specifies any location by clicking, etc.
For example, in the summary example of Fig. 24, for the expression "A quote for Hawaii was provided," a candidate such as "A quote for Hawaii was mailed the other day" is presented. This correction suggestion may have a function for accepting feedback if the suggestion is incorrect or not very good. For example, a "check mark" indicating that the suggestion is accepted or a "cross mark" indicating that the suggestion is not accepted may be displayed below the correction suggestion, and a selection from the user may be accepted.
Alternatively, instead of accepting or rejecting a correction suggestion, a "good" or "bad" mark may be displayed to receive an evaluation of whether the suggestion is good or bad, and evaluation from the user may be received.
When a character string such as a word is modified, the modified character string may be input again into the machine learning model, and the entire summary or the summary following the modified character string may be regenerated by the machine learning model.

The following are examples of UIs (User Interfaces) that are easy for people to edit for a single summary:
- A UI is provided that allows the user to compare and edit the summary output by AI and the text of the revised summary. The feedback module 217 accepts edits to the summary generated by the summary generation module 214. The summaries before and after the revision can be displayed in a form that allows comparison. For example, the summaries before and after the revision can be displayed by switching between tabs, or displayed side by side. The edited parts can also be highlighted by changing the color, etc.
Provides a function for modifying the summary generated by the summary generation module 214 using templates.
-Provide a function to compare the outputs of multiple summaries and select one. Note that a configuration in which the outputs of multiple machine learning models can be compared and selected may be provided. Note that a configuration in which the outputs of multiple machine learning models can be compared and selected may be provided.

FIG. 26 is an example of a summary setting screen 2600.
This is a screen for setting the output of a summary. The summary setting screen is displayed by selecting the summary setting button 2436 in Fig. 24 or the summary setting button 2536 in Fig. 25. The summary setting screen 2600 may be displayed as a separate screen, may be displayed in parallel with another UI, or may be displayed in the form of an overlay (whole or part).
26 shows an example of a summary setting screen that is displayed when "insurance" has already been selected as the case category. Insurance type 2601 is displayed as a subcategory, and automobile insurance has been selected.
As a sub-subcategory, a response method type 2602 is displayed, and procedure is selected.

The keyword specification 2603 specifies the terms to be used when generating a summary. The summary generation module 214 generates a summary including the terms specified here.
Length adjustment 2604 adjusts the length of the summary to be generated, and in this example, a summary of up to 180 characters is generated. Note that, although an upper limit for the length of the summary is specified in this example, a lower limit may also be specified, or a guideline for the length of the summary may also be specified.
Selecting bullet points 2605 generates a summary of bullet points rather than long sentences.

When the function for correcting variations in date and time expressions 2606 is selected, even if various formats of date and time expressions are used in the dialogue text, a summary is generated in which the date and time are expressed in a unified format.
In addition, the system may have a function for correcting variations in the spelling of expressions. For example, a setting can be made to correct all instances of "kyo" to "kyoto".
The summary portion specification 2607 can specify the location of input of the dialogue text, memos, etc. to be summarized. For example, by clicking on the summary portion specification 2607, a screen for accepting specification of a partial area of the dialogue text or memos is displayed, and the specified area can be used as the summary target.
By configuring these settings and pressing the summary start button 2435, the summary generation module 214 generates a summary from the dialogue text with the configured contents.

As mentioned above, we provide a variety of features to vary the summary output:
- A function to control the output of summaries by specifying category information of the inquiry content.
- A function that allows you to specify keywords and have those words reflected in the output.
- Ability to adjust summary length.
- A function to output summaries in bullet point format.
- A function to correct variations in the spelling of expressions.
- A function to specify the summary section from input such as dialogue text or notes.

FIG. 27 shows an example of a reference information call setting screen 2700.
When the customer response processing module 211 accepts the selection of the dialogue text button 2701 at the top left of the screen, it displays the contents of the dialogue text.
When the customer response processing module 211 accepts the selection of the FAQ button 2702, it calls the data search module 218. The data search module 218 acquires the category information that has already been selected, and acquires the FAQ data specified by the corresponding category ID etc. 1202 from the FAQ information 1200 in FIG.
In addition, it is also possible to display FAQs used during customer service, FAQs searched based on the dialogue text, etc.

As mentioned above, the category information can be specified by the operator, but it can also be automatically selected from the contents of the dialogue text. Therefore, when the operator selects the FAQ button 2702, the related FAQ data associated with these dialogue texts is obtained and displayed via the category information.

External link 2703 is a button for calling other external systems.
In addition to the dialogue text, FAQs referred to during the call, logs of other services used such as Web systems, etc. can also be used as input for generating a summary.
FIG. 28 shows an example of a feedback reception screen 2800.
The feedback module 217 receives a rating 2801 for the generated summary. The feedback module 217 may also be referred to as a rating acquisition module.
The feedback module 217 can also accept edits or modifications from the user to the summary displayed in the summary text field 2830. Note that instead of the feedback module 217, a configuration may be provided that includes a separate summary editing module that accepts edits to the summary.

The feedback module 217 accumulates highly rated summary sentences, the document information (dialogue text) on which the highly rated summary sentences are based, and summary sentences revised by the user, and can use these to retrain the machine learning model and improve the machine learning model.
In addition, for example, a database can store the summary generated by the AI, the summary revised by the user, the evaluation score, and an indicator of the difference between the user's summary and the AI's summary.

When the text modified by the user is reflected in CRM or the like using API, the saving is reflected at the time of the operation, and when pasting separately using copy and paste or the like, the saving can be reflected when the copy button is pressed. Also, instead of the copy button, a button for update or reflection may be used. Also, updating may be performed automatically in the background, without the need to explicitly press a button.
In this way, the accuracy of the model can be improved by using a human-in-the-loop mechanism in which the user repeatedly evaluates and corrects the generated summary.
In addition, relearning can use the score of the difference between a summary generated by AI and a summary revised by a human, as well as information such as how much the revisions have changed.

In addition to manually retraining a model on a regular basis, a pipeline can be built to perform retraining on a regular basis.
As for the data to be learned, in addition to a method of relearning using all data in the database, a method of preferentially learning data in the database that did not have very good scores can also be considered.
It is also possible to provide a system in which correction work for the results of model generation can be done through BPO (Business Process Outsourcing) and the results can be improved before being provided.

FIG. 31 shows another example of a feedback acceptance screen 3100.
The user can output the template and start writing a new summary from there. For example, when the feedback module 217 receives the selection of the "template output" button 3131 displayed next to the summary 3130, the feedback module 217 displays a template for optional input to the summary. The feedback module 217 can change the content of the template to be output based on the content of the category (procedural, automobile insurance, etc.) previously input.

Furthermore, in cases where category information has not been input in advance, the feedback module 217 may display a template selection panel 3132 in accordance with the selection of the template output, and may accept the selection of the template within this panel. Also, categories may be in a hierarchical structure, and a template may be automatically selected based on the contents of the dialogue, or a candidate with a high probability may be recommended.
The summary generation module 214 may be configured to output a template.

FIG. 29 shows an example of a work status display screen 2900 displayed on the manager terminal 103 .
The user terminal management module 410 of the administrator terminal 103 provides a function that can monitor the efficiency of ACW work of multiple users who use each user terminal.
The user terminal management module 410 displays distribution and statistics of the time spent in ACW for each important attribute. For example, the ACW work log for each user can be displayed while sorting and filtering by attributes such as time (month, week, etc.), user (operator), and customer service content category.
In the example of Fig. 29, the list of each call (interaction) can be confirmed and filtered. When each item is selected, the user terminal management module 410 displays a work status details display screen 3000 of Fig. 30.

FIG. 30 shows an example of a work status details display screen 3000.
On the work status detail display screen 3000, the dialogue text field 3010 displays text information of the dialogue text, and the memo field 3020 displays memos entered by the user.
Moreover, information on the selected category is displayed in an inquiry content input field 3050, and the selected category is confirmed by pressing a set button 3051.
The summary section 3030 displays a summary generated from the dialogue text.
If the category selection is corrected, the summary can be generated again by pressing the summary start button 2031 again.

When the "reflect in CRM" button 3060 is pressed, the entire content displayed here is finally stored in the CRM system.
According to this embodiment, it is possible to improve the efficiency of ACW work, which requires a large number of man-hours. Also, a highly accurate summary is automatically generated, and it is possible to display it immediately on the Web.
Furthermore, the generated summary can be displayed in an easy-to-edit UI, making it possible to carry out the entire ACW process with a single simple customer response management system 1.

The present invention is not limited to the above-described embodiments, but includes various modified examples. For example, the above-described embodiments have been described in detail to clearly explain the present invention, and are not necessarily limited to those having all of the configurations described. It is also possible to replace part of the configuration of one embodiment with the configuration of another embodiment, and it is also possible to add the configuration of another embodiment to the configuration of one embodiment. It is also possible to add, delete, or replace part of the configuration of each embodiment with other configurations.

Furthermore, each of the above configurations, functions, processing units, processing means, etc. may be realized in hardware, in part or in whole, for example by designing them as integrated circuits. Furthermore, each of the above configurations, functions, etc. may be realized in software by a processor interpreting and executing a program that realizes each function. Information on the programs, tables, files, etc. that realize each function can be stored in a memory, a recording device such as a hard disk or SSD (Solid State Drive), or a recording medium such as an IC card, SD card, or DVD.

In addition, the control lines and information lines shown are those that are considered necessary for the explanation, and not all control lines and information lines in the product are necessarily shown. In reality, it can be considered that almost all components are connected to each other.
The above-described embodiments disclose at least the configurations described in the claims.

1...customer response management system, 102...user terminal, 103...administrator terminal, 211...customer response processing module, 212...category selection module, 213...document information acquisition module, 214...summary generation module, 215...memo information acquisition module, 216...highlighting module, 217...feedback module, 218...data search module

Claims (39)

A document information acquisition means for acquiring first document information;
a summary generating means for inputting the first document information acquired by the document information acquiring means into a trained machine learning model that has performed machine learning to generate a summary using document information as an input, and acquiring a summary for the first document information from the machine learning model;
an output processing means for outputting the summary obtained by the summary generating means;
A document management system having: 2. The document management system according to claim 1, further comprising a category specifying means for specifying a category to which the first document information belongs. The document management system of claim 2, wherein the summary generation means selects a second machine learning model corresponding to the category and inputs the first document information into the second machine learning model, thereby obtaining a summary for the first document information from the second machine learning model. The document management system according to claim 3 , wherein the second machine learning model is a trained machine learning model that has been subjected to machine learning using a plurality of pieces of document information belonging to the category as training data. The document management system of claim 2, wherein the summary generation means inputs information indicating the category and the first document information to obtain a summary related to the category for the first document information from the machine learning model. The document management system according to claim 5 , wherein the machine learning model is a trained machine learning model that has been subjected to machine learning using information indicating a category to which document information belongs and a plurality of document information belonging to the category as training data. 3. The document management system according to claim 2, wherein said summary generating means outputs a template of a summary related to said category. 8. The document management system according to claim 1, further comprising summary editing means for accepting edits to the summary. 9. The document management system according to claim 8, wherein the summary generating means inputs a part of the edited summary and regenerates at least a part of the summary. 9. The document management system according to claim 8, wherein said summary editing means outputs a template for modifying said summary. The document management system according to any one of claims 1 to 7, comprising an evaluation acquisition means for receiving an evaluation of the summary. The document management system according to claim 11, wherein the machine learning model is retrained using document information from which the received highly-rated summary was generated. the summary generating means receives input of setting values for the summary to be generated;
8. The document management system according to claim 1, wherein output contents of the generated summary are changed in accordance with the set value. A document management system according to any one of claims 1 to 7, wherein the summary generation means inputs second document information related to a first document acquired by a second information acquisition means and the first document information, and generates a summary. 8. The document management system according to claim 1, further comprising highlighting means for highlighting one or more characteristic character strings included in the summary that satisfy a predetermined condition. the predetermined condition is the presence of information regarding attributes including at least one of a date, a time, a name, a location, a telephone number, a subscriber number, a customer number, an address, a quantity, and an address;
16. The document management system according to claim 15, wherein said highlighting means highlights a character string including information related to said attribute, out of one or more of said characteristic character strings included in said summary. the predetermined condition is that the characteristic character string is present in the abstract but not in the first document information;
16. The document management system according to claim 15, wherein the highlighting means highlights, of the one or more characteristic character strings included in the summary, a characteristic character string that is present in the summary but not in the first document information. the predetermined condition is that a corresponding characteristic character string exists between the summary and the first document information;
16. The document management system according to claim 15, wherein the highlighting means highlights, of the one or more characteristic character strings included in the summary, a characteristic character string that corresponds between the summary and the first document information. the predetermined condition is that a character string that may possibly result in a low accuracy of the summary generated by the summary generating means is present in the summary;
The document management system according to claim 15, wherein the highlighting means highlights, among one or more of the characteristic character strings contained in the summary, a character string that may be a low-accuracy summary in the summary generated by the summary generating means. A document management system according to any one of claims 1 to 7, wherein the output processing means outputs the summary obtained by the summary generation means and the first document information obtained by the document information acquisition means in association with each other. The document management system according to claim 20, further comprising a highlighting means for highlighting one or more characteristic character strings contained in at least one of the summary and the first document information that satisfy a predetermined condition. The predetermined condition is that there is information regarding an attribute including at least one of a date, a time, a name, a place, an address, a quantity, and an address;
22. The document management system according to claim 21, wherein the highlighting means highlights a character string including information related to the attribute, out of one or more of the characteristic character strings included in at least one of the summary and the first document information. the predetermined condition is that the characteristic character string is present in the abstract but not in the first document information;
22. The document management system according to claim 21, wherein said highlighting means highlights, of the one or more characteristic character strings included in said summary, a characteristic character string that is present in said summary but not in said first document information. the predetermined condition is that a corresponding characteristic character string exists between the summary and the first document information;
22. The document management system according to claim 21, wherein the highlighting means highlights one or more of the characteristic character strings contained in at least one of the summary and the first document information that correspond between the summary and the first document information. the predetermined condition is that a character string that may cause the summary generated by the summary generating means to have a low accuracy is present;
22. The document management system according to claim 21, wherein the highlighting means highlights one or more of the characteristic character strings contained in at least one of the summary or the first document information, which character string may be a character string that is likely to be a low-accuracy summary in the summary generated by the summary generating means. A document management system according to any one of claims 1 to 7, wherein the output processing means outputs the summary acquired by the summary generation means and the second document information acquired by the second information acquisition means in association with each other. The document management system according to claim 26, further comprising a highlighting means for highlighting one or more characteristic character strings contained in at least one of the summary and the second document information that satisfy a predetermined condition. The predetermined condition is that there is information regarding an attribute including at least one of a date, a time, a name, a place, an address, a quantity, and an address;
28. The document management system according to claim 27, wherein said highlighting means highlights a character string including information related to said attribute, among one or more of said characteristic character strings included in at least one of said summary and said second document information. the predetermined condition is that the characteristic character string is present in the second document information but not in the summary;
28. The document management system according to claim 27, wherein the highlighting means highlights, of the one or more characteristic character strings contained in the second document information, a characteristic character string that is present in the second document information but not in the summary. the predetermined condition is that a corresponding characteristic character string exists between the summary and the second document information;
28. The document management system according to claim 27, wherein the highlighting means highlights one or more of the characteristic character strings contained in at least one of the summary and the second document information that correspond between the summary and the second document information. the predetermined condition is that a character string that may possibly result in a low accuracy of the summary generated by the summary generating means is present in the summary;
28. A document management system as described in claim 27, wherein the highlighting means highlights one or more of the characteristic character strings contained in at least one of the summary or the second document information, which character string may be a character string that is likely to be a low-accuracy summary in the summary generated by the summary generating means. A document management system according to any one of claims 1 to 7, wherein the output processing means outputs the first document information acquired by the document information acquisition means, the summary acquired by the summary generation means, and the second document information acquired by the second information acquisition means in association with each other. A second information acquisition means for acquiring second document information,
33. The document management system according to claim 32, further comprising a highlighting means for highlighting one or more characteristic character strings contained in at least one of the summary, the first document information, and the second document information, the characteristic character string satisfying a predetermined condition. the predetermined condition is that the characteristic character string corresponding to the second document information and the summary exists;
34. The document management system according to claim 33, wherein the highlighting means highlights the characteristic character string corresponding to the second document information and the summary among one or more characteristic character strings contained in at least one of the first document information, the second document information, and the summary. the predetermined condition is that the characteristic character string is present in the second document information but not in the summary;
34. A document management system as described in claim 33, wherein the highlighting means highlights one or more of the characteristic character strings contained in at least one of the first document information or the second document information, the characteristic character strings being present in the second document information but not in the summary. the predetermined condition is that the characteristic character string is present in the abstract but not present in the first document information or the second document information;
34. The document management system according to claim 33, wherein the highlighting means highlights one or more of the characteristic character strings contained in the summary that are present in the summary but not present in the first document information or the second document information. The document information acquisition means receives input of voice information,
8. The document management system according to claim 1, wherein the first document information is acquired by analyzing the voice information and converting it into text. A program for causing a management server to function as each means of a document management system according to any one of claims 1 to 37. Obtaining first document information;
inputting the first document information into a trained machine learning model that has performed machine learning to generate a summary using document information as an input, thereby obtaining a summary for the first document information from the machine learning model;
The document management method outputs the summary.

Images