WO2017076205A1 - Method and apparatus for obtaining reply prompt content for chat start sentence - Google Patents

Abstract

Disclosed are a method and apparatus for obtaining reply prompt content for a chat start sentence. In the method, by creating topic databases corresponding to preset topics and obtaining a topic classification to which a chat start sentence received by a communication terminal belongs; performing semantic match on the chat start sentence by utilizing a topic database corresponding to a preset topic that is the same as the topic classification, to obtain a first semantic match result; and if the first semantic match result is not obtained, performing data acquisition on user network data of the communication terminal based on a distributed cloud computing method, and performing semantic match on the chat start sentence by utilizing the user network data, to obtain a second semantic match result, the technical problems of low degree of intelligence and poor user experience of a chat because the conventional database match method cannot certainly obtain chat reply prompt content that matches a chat start sentence are resolved, the accuracy of obtaining reply prompt content is improved, a relatively high degree of intelligence is reflected, and user experience is improved.

Description

Method and device for obtaining reply prompt content of chat initiation sentence Technical field

The present invention relates to the field of communications technologies, and in particular, to a method and apparatus for obtaining a reply prompt content of a chat initiation sentence.

Background technique

At present, whether it is the smart chat reply prompt content provided by the smart chat robot system, or the chat reply prompt content provided by the communication terminal to the communication parties for selection or intelligent reply, most of them are obtained through database matching. The method for obtaining the content of the chat reply prompt by the database matching method is mainly divided into two steps, that is, the pre-processing of the chat initiation sentence is first obtained, the word segmentation text is obtained, and then the obtained segmentation text is matched with the pre-established database to obtain Chat reply to the prompt content.

However, the traditional database matching method does not necessarily obtain the chat reply prompt content matching the chat initiation sentence, thereby resulting in low intelligence and poor user experience.

Summary of the invention

The present invention provides a method and apparatus for obtaining a reply prompt content of a chat initiation sentence, so as to solve the problem that the traditional database matching method does not necessarily obtain the chat reply prompt content matching the chat initiation sentence, resulting in low intelligence of the chat and Technical problems with poor user experience.

According to an aspect of the embodiments of the present invention, a method for obtaining a reply prompt content of a chat initiation sentence is provided, including:

Establish a topic database corresponding to the preset topic;

Obtaining a topic classification to which the chat initiation sentence received by the communication terminal belongs;

Securitally matching the chat initiation sentence by using the topic database corresponding to the preset topic of the topic classification, obtaining the first semantic matching result, and using the first semantic matching result as the reply prompt content of the chat initiation sentence;

If the first semantic matching result is not obtained, the user network data of the communication terminal is collected based on the distributed cloud computing manner, and the chat initiation sentence is semantically matched by using the user network data to obtain the second semantic matching result, and the first The second semantic matching result is used as the reply prompt content of the chat initiation sentence.

Optionally, establishing a topic database corresponding to the preset topic includes:

Setting a scene entry associated with the preset topic, and a scene option corresponding to the scene entry;

A sample chat pair with a preset topic as a chat topic is created, and the sample chat pair is used as a topic database corresponding to the preset topic, and the sample chat pair includes a sample initiation sentence, and a sample response sentence corresponding to the sample initiation sentence set according to the scene option. .

Optionally, obtaining the topic classification to which the chat initiation sentence received by the communication terminal belongs includes:

Obtaining the above chat content of the chat initiation sentence, and merging the chat content of the chat initiation sentence and the chat initiation sentence into a combined text in a text format;

Extract keywords of the merged text;

The topic classification to which the chat initiation sentence belongs is obtained according to the keyword.

Optionally, the session initiation sentence is semantically matched by using a topic database corresponding to the preset topic of the topic classification, and obtaining the first semantic matching result includes:

Obtaining a topic database corresponding to the preset topic that is the same as the topic classification;

Collecting content information of a scene item associated with the topic classification to obtain scene information;

The same sample initiation sentence as the chat initiation sentence is matched in the topic database, and the first semantic matching result is obtained according to the scenario information.

Optionally, semantically matching the chat initiation sentence by using the user network data, and obtaining the second semantic matching result includes:

The user network data is preprocessed to obtain preprocessed text, and the preprocessing includes word segmentation processing, semantic disambiguation processing, part of speech labeling processing, removal of stop word processing, punctuation symbol processing, and expression character processing;

The K-means clustering algorithm is used to perform text clustering on the pre-processed text to obtain a text clustering center;

Extracting keywords of the text clustering center as clustering topics corresponding to the text clustering center;

Obtaining a clustering topic that is closest to the topic classification to which the chat initiation sentence belongs;

The chat initiation sentences are matched in the user network data corresponding to the clustering topic, and the second semantic matching result is obtained.

Optionally, the scene entry includes:

Relationship entry, name entry, gender entry, age entry, instant messaging account entry, email address entry, home address entry, occupation category entry, job entry, work unit entry, unit address entry of the communication terminal that sends and receives the chat initiation sentence , bank account entry, friend impression entry, hobbyist entry, circle of friends status entry, mood entry, recent topic of interest entry, current communication status entry, scene image entry, time entry, holiday entry, season entry, geographic location information entry, distance An entry, a communication frequency entry, a communication time entry, a communication duration entry, and a selection mode entry for initiating historical communication, wherein the selection manner includes starting a communication mode from the address book, initiating a communication mode from the historical call record, and initiating a communication mode from the short message communication module, Initiate communication from the dial pad.

Optionally, collecting the content information of the scene image entry of the communication terminal that sends or receives the chat initiation sentence in the scene entry includes:

Collecting a scene image of a communication terminal that sends or receives a chat initiation sentence;

The Gaussian difference DOG operator is used to extract the region of interest of the scene training image, and the scale-invariant feature transform SIFT feature of the region of interest of the scene training image is calculated;

The K-means clustering algorithm is used to cluster the SIFT features of the region of interest of the scene training image to obtain a plurality of cluster centers, and a visual word dictionary composed of visual words corresponding to each cluster center is established;

The DOG operator is used to extract the region of interest of the scene image, and the visual word dictionary is matched with the visual word closest to the SIFT feature of the region of interest of the scene image;

According to the distribution of the visual words of the region of interest of the scene image, the scene image is classified by using a pre-trained support vector machine classifier, and the communication for sending or receiving the chat initiation sentence is obtained. The content information of the scene image entry of the terminal.

According to another aspect of the embodiments of the present invention, an apparatus for obtaining a reply prompt content of a chat initiation sentence is provided, including:

a topic database creation device, configured to establish a topic database corresponding to the preset topic;

The topic classification obtaining device is configured to obtain a topic classification to which the chat initiation sentence received by the communication terminal belongs;

The first semantic matching device is configured to perform semantic matching on the chat initiation sentence by using a topic database corresponding to the preset topic of the topic classification, obtain a first semantic matching result, and use the first semantic matching result as a reply prompt of the chat initiation sentence. content;

The second semantic matching device is configured to collect data of the user network data of the communication terminal based on the distributed cloud computing manner if the first semantic matching result is not obtained, and perform semantic matching on the chat initiation sentence by using the user network data to obtain the first The second semantic matching result, and the second semantic matching result is used as the reply prompt content of the chat initiation sentence.

Optionally, the topic database creation device includes:

a setting device configured to set a scene item associated with the preset topic and a scene option corresponding to the scene item;

The sample chat pair creation device is configured to create a sample chat pair with a preset topic as a chat topic, and the sample chat pair is used as a topic database corresponding to the preset topic, and the sample chat pair includes a sample initiation sentence, and is set according to the scene option. The sample response sentence corresponding to the sample initiation sentence.

Optionally, the topic classification obtaining device includes:

a merged text obtaining device, configured to obtain the above chat content of the chat initiation sentence, and merge the chat start sentence and the chat content of the chat initiation sentence into a merged text in a text format;

a keyword extracting device configured to extract keywords of the merged text;

The topic classification determining means is configured to acquire a topic classification to which the chat initiation sentence belongs according to the keyword.

Embodiments of the present invention have the following beneficial effects:

The method and device for obtaining the reply prompt content of the chat initiation sentence provided by the embodiment of the present invention, the method for obtaining the topic classification to which the chat initiation sentence received by the communication terminal belongs, and using the customized preset topic corresponding to the topic classification The topic database semantically matches the chat initiation sentence, obtains the first semantic matching result, collects the user network data of the communication terminal without obtaining the first semantic matching result, and uses the user network data to semantically match the chat initiation sentence. The second semantic matching result is obtained, which solves the problem that the traditional database matching method does not necessarily obtain the chat reply prompt content matching the chat initiation sentence, thereby causing low intelligence of the chat and technical problems of poor user experience, fully The user network data of the communication terminal is used to obtain the reply prompt content of the chat initiation sentence, which improves the accuracy of the reply prompt content acquisition, reflects a high level of intelligence, and improves the user experience.

In addition to the objects, features and advantages described above, the present invention has other objects, features and advantages. The invention will now be described in further detail with reference to the drawings.

DRAWINGS

The accompanying drawings, which are incorporated in the claims In the drawing:

1 is a flowchart of a method for obtaining a reply prompt content of a chat initiation sentence according to an optional embodiment of the present invention;

2 is a scenario image of a communication terminal that assumes a received chat initiation sentence according to an alternative embodiment of the present invention;

3 is a diagram of a visual word result obtained by matching a scene image of a communication terminal of a received chat initiation sentence with a visual word dictionary according to an alternative embodiment of the present invention;

4 is a flowchart of a method for obtaining a reply prompt content of a chat initiation sentence according to an alternative embodiment of the present invention;

FIG. 5 is a flowchart of a method for obtaining a reply prompt content of a chat initiation sentence according to an alternative embodiment of the present invention;

6 is a flowchart of a method for obtaining a reply prompt content of a chat initiation sentence for a third simplified embodiment according to an alternative embodiment of the present invention;

FIG. 7 is a structural block diagram of an apparatus for obtaining a reply prompt content of a chat initiation sentence according to an optional embodiment of the present invention.

detailed description

The embodiments of the present invention are described in detail below with reference to the accompanying drawings.

Referring to FIG. 1, an alternative embodiment of the present invention provides a method for obtaining a reply prompt content of a chat initiation sentence, including:

Step S101, establishing a topic database corresponding to the preset topic;

Step S102, acquiring a topic classification to which the chat initiation sentence received by the communication terminal belongs;

Step S103: Perform semantic matching on the chat initiation sentence by using a topic database corresponding to the preset topic of the topic classification, obtain a first semantic matching result, and use the first semantic matching result as a reply prompt content of the chat initiation sentence;

Step S104: If the first semantic matching result is not obtained, collect data of the user network data of the communication terminal based on the distributed cloud computing manner, and perform semantic matching on the chat initiation sentence by using the user network data to obtain a second semantic matching result. And the second semantic matching result is used as the reply prompt content of the chat initiation sentence.

The method for obtaining the reply prompt content of the chat initiation sentence provided by the embodiment of the present invention obtains the topic classification to which the chat initiation sentence received by the communication terminal belongs, and uses the customized topic database corresponding to the preset topic with the same topic classification to chat. Initiating a sentence for semantic matching, obtaining a first semantic matching result, and collecting user network data of the communication terminal without obtaining the first semantic matching result, and using the user network data to semantically match the chat initiation sentence to obtain a second The semantic matching result solves the problem that the traditional database matching method does not necessarily obtain the chat reply prompt content matching the chat initiation sentence, thereby resulting in low intelligence of the chat and poor user experience, and fully utilizing the communication terminal. User network data to get chat The reply prompt content of the sentence improves the accuracy of the content of the reply prompt content, embodies a higher level of intelligence, and improves the user experience.

The user network data of the communication terminal in this embodiment includes personal information data of the communication terminal, social information data (microblog, WeChat, forum, blog, etc.), communication information data, online shopping information data, online footprint information data, and the like. . The communication information includes the user's own historical communication records, historical communication records of other users using the same communication application software, and communication records provided by third-party application software. Optionally, the communication record further includes a call record and a short message record, and the short message record further includes a mobile phone short message record and an instant message record, and the call record further includes a mobile phone call record and an instant message voice and video call record. It should be noted that, in this embodiment, the reply prompt content matching the chat initiation sentence in the chat process is mainly obtained based on the user network data, so the embodiment is mainly for the network chat data with context interaction in the user network data of the communication terminal. Collecting, such as WeChat, QQ instant messaging chat, chat data with Taobao customer service, Baidu question and answer, interaction in Weibo private message or chat data. In addition, in the actual implementation process, the present embodiment can perform semantic matching on the chat initiation sentence by using the topic database corresponding to the preset topic with the same topic classification, and can also use the topic corresponding to the preset topic closest to the topic classification. The database semantically matches the chat initiation sentence.

This embodiment uses a Hadoop-based platform for processing big data. Hadoop is an open source distributed computing platform, and its core includes the distributed file system (Hadoop Distributed Files System, Hadoop, referred to as HDFS). The many advantages of HDFS (mainly including high fault tolerance, high scalability, etc.) allow users to deploy Hadoop on low-cost hardware and build distributed clusters to form a distributed system. Hadoop DataBase (HBase) is a distributed database system based on the distributed file system HDFS that provides high reliability, high performance, column storage, scalable, real-time read and write. It is mainly used to store unstructured data. Loose and semi-structured loose data. In this embodiment, the collected network data is stored by the distributed storage device, and the distributed storage device is implemented based on HDFS.

Optionally, establishing a topic database corresponding to the preset topic includes:

Setting a scene entry associated with the preset topic, and a scene option corresponding to the scene entry;

A sample chat pair with a preset topic as a chat topic is created, and the sample chat pair is used as a topic database corresponding to the preset topic, and the sample chat pair includes a sample initiation sentence, and a sample response sentence corresponding to the sample initiation sentence set according to the scene option. .

Since the phrase for the same chat (for example, "Go travel together?") often requires different responses in real life (for example, "The weather is bad, how about going next time, how?", "I prefer the house." At home, I don't like to travel.", "Recent work is too busy, I can't get time to travel." etc.), that is, for the same chat initiation sentence, communication terminal users often need to give different according to different environments or scenarios. Reply. To solve the problem, in this embodiment, when establishing a topic database corresponding to a preset topic, first setting a scene item associated with the preset topic, and a scene option corresponding to the scene item, and then creating a topic with the preset topic as the chat topic. The sample chat pair includes a sample initiation sentence, a sample response sentence corresponding to the sample initiation sentence set according to the scene option, and a sample chat pair with the preset topic as the chat topic as a topic database corresponding to the preset topic.

Optionally, the embodiment sets the scene item associated with the preset topic according to daily experience. Referring to Table 1, Table 1 shows several different preset topics and their associated scene entries. For example, if the preset topic is a "tourism" default topic, it is known from daily experience that when the communication terminal replies to the chat initiation sentence belonging to the "tourism" preset topic, it is necessary to take into consideration the factors of the weather, There are no time factors, location factors, and whether there are tourism interest factors, etc., and when the default topic is "email" preset topic, it is known through daily experience that the communication terminal belongs to the "mailing" preset topic. When replying to a chat initiation sentence, it is necessary to take into account the e-mail address factor sent or received by the e-mail, and basically does not consider the factors of the weather. It should be noted that the embodiment sets the preset topic for different topics. The scene entries associated with them are not fixed, but are only given according to daily experience, that is, the user can set the scene entries associated with the preset topics as needed.

Table 1

In this embodiment, the scene option corresponding to the scene entry is also customized by the user as needed. For example, when setting scene options corresponding to weather items, you can include three scene options ("1" for sunny, "2" for rain, "3" for snow), or just two scene options ( “1” means the weather is good, “2” means the weather is bad); similarly, when setting the scene option corresponding to the time entry, it can include three scene options (“1” means morning, “2” means afternoon, “3” "Indicating evening", it is also possible to include only two scene options ("1" for daytime and "2" for nighttime).

It should be noted that, in this embodiment, the sample corresponding to the sample initiation sentence is set according to the scene option. In this reply sentence, it is necessary to fully consider the combination of all scene options of all scene entries associated with the preset topic. Referring to Table 2, Table 2 is the code name of the scene option corresponding to the three scene entries associated with the "shopping" preset topic, wherein the relationship entry may refer to the relationship entry of the communication terminal that receives and sends the chat initiation sentence. It can be seen from Table 2 that the scene option corresponding to the relationship entry is six items, the scene option corresponding to the distance item is three items, and the scene option corresponding to the weather item is five items. Therefore, the number of combinations of all scene options of all scene entries in this embodiment is 6*3*5, that is, when creating a sample chat pair in the topic database corresponding to the “shopping” preset topic, a sentence is initiated for each sample. You can set up to 90 sample response sentences corresponding to it.

In this embodiment, when creating a sample reply sentence for a sample start sentence, the number of combinations of the scene options and the sample reply sentence of the combined mode may be set as needed, that is, the start sentence for each sample does not need to set all the scene option combinations. Sample response sentence. And in an optional implementation process, the embodiment sets the content information for each scene item to include a scene option that is “empty” (which may be represented by a “0” code), because in actual implementation, The content information corresponding to the scene entry cannot be obtained. For example, if the communication terminal that receives the chat initiation sentence does not have a Global Positioning System (GPS) positioning or does not open the geographic location permission, the data returned by the system is air. For example, in some cases where it is necessary to rely on the Internet for querying and searching to obtain the content information of the scene item, the related search result cannot be obtained when the communication terminal disconnects the network link or enters the no-network signal area. In addition, when creating a sample chat pair in the topic database, for example, for some chat initiation sentences with a fixed reply, the communication terminal user only needs to set the scene option content to be empty.

Table 2

In this embodiment, by setting a scene item associated with a preset topic and a scene option corresponding to the scene item, and creating a sample chat pair with the preset topic as a chat topic as a topic database corresponding to the preset topic, Enrich the sample reply sentence types for the same sample-initiating sentence, meet the actual needs, enhance the user experience, and realize the sample setting of the sample-initiating sentence from different scenes and their combination conditions by considering the scene items associated with the preset topic. The reply sentence is in line with the logical thinking of human replying to chat information, and has a high level of intelligence and personalization.

Optionally, obtaining the topic classification to which the chat initiation sentence received by the communication terminal belongs includes:

Obtaining the above chat content of the chat initiation sentence, and merging the chat content of the chat initiation sentence and the chat initiation sentence into a combined text in a text format;

Extract keywords of the merged text;

The topic classification to which the chat initiation sentence belongs is obtained according to the keyword.

The embodiment determines that the topic classification to which the chat initiation sentence belongs is not only based on the chat initiation sentence but on the above chat content of the chat initiation sentence and the chat initiation sentence, but in the actual implementation process, the chat initiation sentence is relative to the chat. The above-mentioned chat content of the initiating sentence has more reference value for obtaining the topic classification to which the chat initiation sentence belongs. Therefore, in this embodiment, the keyword of the merged text can be extracted by using the weighted word frequency statistics of the segmentation text obtained after the merged text segmentation, that is, The closer the chat content is to the chat initiation sentence, the greater the weighting factor. In this embodiment, after the keyword is obtained according to the merged text, the content corresponding to the keyword may be used as the topic classification to which the chat initiation sentence belongs, or the topic corresponding to the keyword may be queried according to the association mapping table of the preset keyword and the topic classification. classification.

In this embodiment, the topic classification of the chat initiation sentence is obtained by combining the chat content of the chat initiation sentence and the chat initiation sentence, and the chat context in which the chat initiation sentence belongs is fully considered, and the chat context is obtained by relying only on the chat initiation sentence. Topic classification is more accurate. And the word frequency statistics of the word segmentation text obtained after the merged text segmentation is determined by the weighting method to determine the keyword, so that the topic classification of the obtained chat initiation sentence is more accurate.

It should be noted that, in this embodiment, when the chat initiation sentence does not chat with the content, the present embodiment only obtains the topic classification to which the chat initiation sentence belongs, and in this embodiment, the above chat content of the chat initiation sentence is obtained. The scope of the user is customized, for example, the above chat content can be obtained within a certain period of time, or the above chat content within a certain number of content.

Optionally, the session initiation sentence is semantically matched by using a topic database corresponding to the preset topic of the topic classification, and obtaining the first semantic matching result includes:

Obtaining a topic database corresponding to the preset topic that is the same as the topic classification;

Collecting content information of a scene item associated with the topic classification to obtain scene information;

The same sample initiation sentence as the chat initiation sentence is matched in the topic database, and the first semantic matching result is obtained according to the scenario information.

In this embodiment, the content information of the scene item collected by the topic classification may be calculated, reasoned, The way to query, search, or any combination of them. Optionally, the method for calculating, inferring, querying, searching, or any combination of personal information, social information, communication information, online shopping information, online footprint information, user behavior information, user service information, and the like of the communication terminal may be The content information corresponding to the scene item is obtained, wherein the user behavior information refers to information about the behavior of demand expression, information acquisition, information utilization, etc., which is displayed when the user seeks the information he needs. The communication information includes the user's own historical communication records, historical communication records of other users using the same communication application software, and communication records provided by third-party application software. Optionally, the communication record further includes a call record and a short message record, and the short message record further includes a mobile phone short message record and an instant message record, and the call record further includes a mobile phone call record and an instant message voice and video call record.

For example, when the scene item is a geographical location information item, it can be obtained by querying the GPS positioning information. When the scene item is a distance item, the geographical position difference of the communication terminal that receives and sends the chat initiation sentence can be obtained by calculating the scene item. When the topic item is recently concerned, it can be obtained by searching the latest web browsing record of the communication terminal. When the scene item is a weather item, it can be obtained by querying the weather webpage, or by reasoning the meteorological information such as temperature, wind direction and humidity collected. .

Obtaining the first semantic matching result according to the scenario information in this embodiment may include: first identifying the scenario information, and obtaining the identifier ID. Optionally, referring to Table 2, it is assumed that only the sending chat initiation sentence and the receiving chat initiation sentence are collected in this embodiment. The content information of the relationship entry of the communication terminal is “colleague”, and the content information of the weather item of the communication terminal that sends the chat initiation sentence is “clear”, and the obtained identification ID number is “3+0+1”, and then The topic database corresponding to the preset topic with the same topic classification matches the same sample initiation sentence as the chat initiation sentence, and matches the combination code corresponding to the identification ID number in the scenario option corresponding to the sample initiation sentence, and will be the same as the identification ID. The sample reply sentence corresponding to the combination code is used as the reply prompt content corresponding to the chat initiation sentence. In an actual implementation process, in this embodiment, when the topic database corresponding to the preset topic with the same topic classification matches the sample initiation sentence with the same chat initiation sentence, the same sample initiation sentence can be obtained by using exact matching to obtain the same sentence initiation sentence. A fuzzy matching method may also be used to obtain a sample initiation sentence similar to a chat initiation sentence. In this embodiment, the fuzzy matching method is used to obtain the chat initiation sentence. The similar sample initiation sentence may include: first preprocessing the chat initiation sentence, the preprocessing includes word segmentation, semantic disambiguation, part of speech tagging, removing the stop word, and the like, and then the pre-processed chat initiation sentence and the topic database The sample initiation sentence performs text matching, and the text matching sample initiation sentence whose similarity is greater than a preset threshold is used as a sample initiation sentence matching the chat initiation sentence. In this way, for the same chat initiation sentence, different sample reply sentences can be matched according to different scene information, thereby realizing that the reply prompt content corresponding to the chat initiation sentence is intelligently acquired according to the scene information of the communication terminal, and has a high degree of intelligence. And personalization level.

It should be noted that the topic database corresponding to the preset topic created in this embodiment has the functions of automatic learning and automatic updating. Optionally, when the content information of the collected scene entry is not included in the created topic database. When the content information of the item is included, for example, when the scene option of the weather item includes only three, respectively: "1" indicates clear, "2" indicates rain, and "3" indicates snowing, and when weather scene entries are collected When the content information is "cloudy", the system will create a scene option with the code "4" indicating "cloudy" under the weather scene entry, and the corresponding updated scene option combination and the corresponding sample reply sentence. In addition, the sample reply sentence that is combined for each scene option in this embodiment may be one session content, or may be multiple session content.

Optionally, semantically matching the chat initiation sentence by using the user network data, and obtaining the second semantic matching result includes:

The user network data is preprocessed to obtain preprocessed text, and the preprocessing includes word segmentation processing, semantic disambiguation processing, part of speech labeling processing, removal of stop word processing, punctuation symbol processing, and expression character processing;

The K-means clustering algorithm is used to perform text clustering on the pre-processed text to obtain a text clustering center;

Extracting keywords of the text clustering center as clustering topics corresponding to the text clustering center;

Obtaining a clustering topic that is closest to the topic classification to which the chat initiation sentence belongs;

The chat initiation sentences are matched in the user network data corresponding to the clustering topic, and the second semantic matching result is obtained.

The second semantic matching result is obtained by matching the chat initiation sentence with the collected user network data, but the user network data is generally the number of big data. According to the information, when the chat initiation sentence is directly matched in the user network data, multiple matching results or the obtained reply prompt content may be completely irrelevant. For this problem, the embodiment firstly collects the user network data. Perform preprocessing, perform text clustering on the preprocessed preprocessed text, obtain the text clustering center, and extract the keywords of the text clustering center as the clustering topic, and finally get the closest to the topic classification to which the chat initiation sentence belongs. The chat initiation sentence is matched in the user network data corresponding to the clustering topic, thereby obtaining the second semantic matching result.

Optionally, the present embodiment performs text clustering on the pre-processed text based on the K-means clustering algorithm to obtain a text clustering center, which may include the following steps:

a. Randomly select K data as a center point, and the center point in this embodiment may be either a word or a sentence.

b. Then calculate the distance between each sample and the center point, and select the center point corresponding to the smallest distance to be the class to which it belongs. Optionally, in this embodiment, the distance between each sample and the center point is obtained by calculating the distance between the word vector corresponding to each sample and the word vector corresponding to the center point.

c. For each class, recalculate the center point (the mean of all samples in the category).

d. Repeat the iterations b and c steps until convergence, that is, the cluster center does not change.

In an optional implementation process, in order to improve the accuracy of obtaining the second semantic matching result according to the user network data, the communication terminal user generally performs one or more screenings on the collected user network data before performing preprocessing and text clustering. . In addition, when obtaining the clustering topic that is closest to the topic classification to which the chat initiation sentence belongs, the embodiment may obtain the similarity between the topic classification and the cluster topic to which the chat initiation sentence belongs, or may be preset. The relevance of topic classification and clustering topic is obtained.

In this embodiment, the K-means clustering algorithm is used to extract the clustering topic of the user network data, and the chat initiation sentence is matched in the user network data corresponding to the clustering topic closest to the chat initiation sentence, thereby saving a large number of user networks. The matching time of the data matching the chat initiation sentence, thereby improving the speed and efficiency of the reply prompt content acquisition, and matching the chat initiation sentence by only the user network data corresponding to the cluster topic closest to the chat initiation sentence, Make the obtained reply prompt content more accurate and intelligent.

Optionally, the scene entry includes:

Name entry, gender entry, age entry, instant messaging account entry, email address entry, home address entry, occupation category entry, job entry, work unit entry, unit address entry, bank account number of the communication terminal that sends and receives the chat initiation sentence Entry, friend impression entry, hobbyist entry, circle of friends status entry, mood entry, recent attention topic entry, current communication status entry, scene image entry, time entry, holiday entry, season entry, geographic location information entry, distance entry, communication Frequency entry, communication number entry, communication duration entry, selection mode entry for initiating historical communication, wherein the selection mode includes starting communication mode from the address book, initiating communication mode from the historical call record, initiating communication mode from the short message communication module, and from the dial pad Initiate communication methods.

The scene entry of this embodiment is not limited to including only the above-mentioned scene entries, and is not limited to including all of the above scene entries, and may be selected by the user or selected according to needs and system design complexity and design precision.

Optionally, collecting the content information of the scene image entry of the communication terminal that sends or receives the chat initiation sentence in the scene entry includes:

Collecting a scene image of a communication terminal that sends or receives a chat initiation sentence;

The region of interest of the scene training image is extracted by using a Difference of Gaussian (DOG) operator, and the SIFT feature of the region of interest of the scene training image is calculated;

The K-means clustering algorithm is used to cluster the Scale-Invariant Feature Transform (SIFT) features of the region of interest of the scene training image to obtain multiple cluster centers, and each cluster is established. a visual word dictionary composed of visual words corresponding to the class center;

The DOG operator is used to extract the region of interest of the scene image, and the visual word dictionary is matched with the visual word closest to the SIFT feature of the region of interest of the scene image;

The scene image is classified according to the distribution of the visual words of the region of interest of the scene image by using a pre-trained support vector machine classifier, and the content information of the scene image item of the communication terminal that transmits or receives the chat initiation sentence is obtained.

The SIFT feature in this embodiment is a scale-invariant feature transformation, which is to find extreme points in the spatial scale and extract its position, scale, and rotation invariants. Optionally, the process of matching the visual word closest to the SIFT feature of the region of interest of the scene image in the visual word dictionary is: the SIFT feature of each region of interest of the scene image and each of the visual word dictionary The SIFT feature of the cluster center corresponding to a visual word is used for similarity calculation. When the similarity between the region of interest and the visual word is calculated to be greater than a preset threshold, the visual word is considered to be the closest visual word to the region of interest. . In this way, all regions of interest of the scene image can be represented by visual words, and the classifier is designed and trained based on the distribution of visual words in the scene image, thereby finally obtaining a scene of the communication terminal that transmits and/or receives the chat initiation sentence. The content information of the image entry.

Referring to FIG. 2, FIG. 2 is a schematic diagram of a scenario in which a communication terminal receiving a chat initiation sentence receives a chat initiation sentence according to an embodiment, and extracts SIFT features of five regions of interest of the scene image, and calculates each sense by calculating The similarity between the SIFT feature of the region of interest and the visual word dictionary, thereby obtaining the closest visual words to the above five regions of interest, namely "sky", "flag", "building", "lion", "bridge" Referring to FIG. 3, the content information of the scene image item of the communication terminal receiving the chat initiation sentence is obtained by using the pre-trained support vector machine according to the obtained visual words closest to the above five regions of interest.

In this embodiment, the content information of the scene image item of the communication terminal is obtained by collecting the scene image of the communication terminal, so that the scene information acquired based on the scene image information is closer to the real scene information, and the scene corresponding to the chat initiation sentence is obtained by using the scene image information. The response prompts are more personalized and more in line with the communication context.

The process and principle of obtaining the reply prompt content of the chat initiation sentence in the embodiment of the present invention are further described below for the three simplified embodiments.

The communication scenario for the first embodiment is as follows: the communication terminal A sends the chat initiation sentence in the text format of “Would you like to go shopping together?” to the communication terminal B. Referring to FIG. 4, the communication terminal B obtains a reply prompt of the chat initiation sentence. Content methods include:

Step S201, setting a scene entry associated with the preset topic, and corresponding to the scene entry Scene options. Optionally, it is assumed that the number of preset topics in the embodiment is sufficient, and the scene entries associated with the “shopping” topic in the preset topic include a relationship entry of the communication terminal that sends and receives the chat initiation sentence, a distance entry, and a receiving chat initiation. The weather entry of the communication terminal of the sentence has 6 scene options corresponding to the relationship entry, 3 scene options corresponding to the distance entry, and 5 scene options corresponding to the weather item, refer to Table 2.

Step S202, creating a sample chat pair with the preset topic as the chat topic, and using the sample chat pair as the topic database corresponding to the preset topic, the sample chat pair including the sample initiation sentence and the sample initiation sentence corresponding according to the scene option. Sample response sentence.

Optionally, the embodiment creates a sample chat pair corresponding to the “shopping” topic, and sets a custom combination number of sample reply sentences for each sample start sentence in each sample chat pair, for example, for the sample reply sentence “Do you want to Go shopping together?” Set the sample response sentence according to all combinations of scene options (90 types). For the sample reply sentence “How do you translate in English?” Set a sample response sentence of the scene option combination (code 0+0+0) Then, the created sample chat pair corresponding to the "shopping" topic is created as a topic database corresponding to the "shopping" topic.

Step S203: Acquire the above chat content of the chat initiation sentence, and merge the chat content of the chat initiation sentence and the chat initiation sentence into a merged text in a text format. Optionally, it is assumed that the content of the chat initiation sentence in the embodiment includes a total of four conversation contents, which may be: communication terminal A: busy? / Communication terminal B: Fortunately. / Communication terminal A: I feel that there is no suitable clothing to wear recently! / Communication terminal B: Yes. Then the merged text obtained is {busy? / Ok. / Recently I feel that there is no suitable clothing to wear! /right. / Do you want to go shopping together? }.

Step S204, extracting keywords of the merged text. After the operation of segmenting the merged text, removing the stop words, part of speech tagging, semantic disambiguation, etc., it is assumed that the extracted word segmentation text includes {"busy", "clothing", "wearing", "shopping"}, and this embodiment The word frequency statistics are performed by weighted word frequency statistics, and the word segmentation text corresponding to the maximum weighted word frequency statistics value is selected as the keyword of the combined text. Optionally, since the merged text of the embodiment includes five pieces of chat content, the five weighting coefficients respectively set are respectively k1=0.5, k2=0.2, k3=0.15, k4=0.1, and k5=0.05, wherein the weighting coefficient The time interval from the chat content in the merged text is from small to large. Decrement, and the specific value of the weighting factor is customized by the user as needed.

Since the word frequency of the word segmentation text extracted in this embodiment is one, after the weight word frequency calculation is performed on each word segmentation text, the keyword can be obtained as “shopping”.

Step S205, determining a topic classification to which the chat initiation sentence belongs according to the keyword. Optionally, in this embodiment, the content corresponding to the keyword is used as the topic classification to which the chat initiation sentence belongs, that is, the topic to which the chat initiation sentence belongs in this embodiment is classified as “shopping”.

Step S206: Acquire a topic database corresponding to the preset topic that is the same as the topic classification.

Step S207: Collect content information of a scene item associated with the topic classification, and obtain scene information. Optionally, the scene entry associated with the “shopping” topic in the preset topic in this embodiment includes: f1=sending a chat initiation sentence and a relationship entry of a communication terminal receiving the chat initiation sentence, f2=sending a chat initiation sentence, and receiving a chat The distance entry of the communication terminal of the initiation sentence, f3 = the weather entry of the communication terminal receiving the chat initiation sentence, see Table 1. And it is assumed that the content information of the relationship entry of the communication terminal that receives the chat initiation sentence and the communication initiation terminal is “friend” by querying the address book remark information of the communication terminal that receives the chat initiation sentence, and is sent by calculation. The location difference between the chat initiation sentence and the communication terminal receiving the chat initiation sentence is f2=the content information of the distance entry of the communication terminal that sends the chat initiation sentence and receives the chat initiation sentence is “far” (pre-defined geographical difference is greater than 10 km) Returning "far" content information to the distance entry), and obtaining the weather information of the geographic location of the communication terminal receiving the chat initiation sentence by searching the Internet for the weather information of the communication terminal receiving the chat initiation sentence is "clear" .

Step S208, matching the same sample initiation sentence as the chat initiation sentence in the topic database, and acquiring the first semantic matching result according to the scenario information. Optionally, the scenario information is first identified, and the identifier ID is obtained. Referring to Table 2, the identifier ID number obtained is “2+1+1”, and then the topic database corresponding to the preset topic with the same topic classification is matched and The same as the sample initiation sentence of the chat initiation sentence, and the combination code corresponding to the identification ID number in the scene option corresponding to the sample initiation sentence, and the sample reply sentence corresponding to the same combination code as the identification ID is corresponding to the chat initiation sentence Reply to the prompt content. It is assumed that the present embodiment initiates the sentence “Do you want to go shopping together” for the sample chat, and the obtained conclusion content corresponding to the reply condition ID number “2+1+1” is 3, respectively, "Pro, I am too far away from you, can you make an appointment next time?", "The weather is good today, I just want to go out and stroll.", "Although I am far away from you, but The weather is good today, so I still want to go out and stroll." In the actual chat reply, the system will provide the three reply prompts for the user to select.

It can be seen that when the content information of the collected scene entries is different, the acquired scene information is different, and the obtained reply prompt content is different, so that the existing method for obtaining the reply prompt content is not combined with the scene information of the communication terminal participating in the chat. The result that the obtained reply prompt content is fixed, the degree of intelligence of the chat is low, and the user experience is not good, and the different reply prompt contents are obtained according to different scene information, and the obtained reply prompt content fully combines the communication parties The scene information is closely related to both parties, reflecting a high level of intelligence and personalization. In addition, the present embodiment obtains the reply prompt content through the topic database without manual editing, which improves the chat session speed and improves the user experience.

The communication scenario for the second embodiment is: the communication terminal A sends the chat initiation sentence in the text format that the content is "busy?" to the communication terminal B. Referring to FIG. 5, the communication terminal B obtains the reply prompt content of the chat initiation sentence. include:

Step S301, setting a scene item associated with the preset topic, and a scene option corresponding to the scene item. Optionally, it is assumed that the number of preset topics in the embodiment is sufficient, and the scene entry associated with the “hello” topic in the preset topic only includes the scene image entry of the communication terminal that receives the chat initiation sentence, and corresponds to the scene image entry. There are a total of nine scene options, namely airports, oceans, forests, villages, streets, trees, high-rises, highways, and offices.

Step S302, creating a sample chat pair with the preset topic as the chat topic, and using the sample chat pair as the topic database corresponding to the preset topic, the sample chat pair including the sample initiation sentence, and the sample initiation sentence corresponding according to the scene option. Sample response sentence.

Optionally, in this embodiment, when creating a sample chat pair with “calling” as a chat topic, a sample reply sentence corresponding to the scene option is set for each sample start sentence in each sample chat pair, for example, for a sample reply sentence “ Where are you?” You can set up sample response sentences (9 types) corresponding to all scene options, or you can set sample responses corresponding to some scene options. sentence.

Step S303: Acquire a topic classification to which the chat initiation sentence received by the communication terminal belongs. Optionally, it is assumed that the content of the chat initiation sentence is empty in the embodiment, and since the content of the chat initiation sentence is simple, the topic to which the chat initiation sentence belongs is classified as “greeting”.

Step S304, acquiring a topic database corresponding to the preset topic that is the same as the topic classification.

Step S305, collecting content information of the scene item associated with the topic classification, and obtaining scene information. Optionally, the scene entry associated with the “greeting” topic in the preset topic in this embodiment includes only f1=the scene image entry of the communication terminal that receives the chat initiation sentence.

Step S306, collecting a scene image of the communication terminal that receives the chat initiation sentence.

Step S307, extracting the region of interest of the scene image, and matching the visual word closest to the SIFT feature of the region of interest of the scene image in the visual word dictionary.

Optionally, in this embodiment, the DOG (Difference of Gaussian) operator is first used to extract the region of interest of the scene image, and then the SIFT feature of each region of interest is calculated, and the region of interest of the scene image is matched in the visual word dictionary. The SIFT features the closest visual word. It is assumed that after matching, the visual words corresponding to the three regions of interest are respectively referred to as “road signs”, “lanes” and “distance signs”.

Step S308, classifying the scene image according to the distribution of the visual words of the region of interest of the scene image by using a pre-trained support vector machine classifier to obtain content information of the scene item. Optionally, the embodiment uses a support vector machine method to design a classifier, and trains 9 types of images classified in a known scene during the training phase (airport, ocean, forest, village, street, tree, tall building, highway, Office), each type of image includes 100 different training sample images. By inputting the scene image ("roadmap", "lane", "distance sign") including three visual words obtained in step S307 to the pre-trained support vector machine classifier, the communication terminal receiving the chat initiation sentence can be obtained. The content information of the scene image entry is "highway".

Step S309, matching the same sample initiation sentence as the chat initiation sentence in the topic database, and acquiring the first semantic matching result according to the scenario information. Optionally, it is assumed that this embodiment is preset In the topic database corresponding to the topic category "Hello", the sample reply sentence for the sample is "Busy?", and the sample response sentence with the scene option "Highway" includes multiple items, for example, "I am at high speed, it is inconvenient to return information. Contact us at the high speed.", "Sorry, it is not convenient to return information, contact you later", etc., then display the contents of these reply prompts in the communication terminal for the user to select.

In this embodiment, the scene image of the communication terminal that receives the chat initiation sentence is collected, and the content information of the scene entry is obtained based on the collected scene image, so that the scene information obtained based on the content information is closer to the real scene information, so that the scene information is adopted. The obtained reply prompt matching the chat initiation sentence is highly intelligent and personalized.

The communication scenario for the third embodiment is as follows: the communication terminal A sends the communication initiation message to the communication terminal B in the text format “What is the price of the apple?”, and the sentence is initiated for the chat, and the embodiment cannot be based on the established topic. The database obtains the first semantic matching result matched by the database. Referring to FIG. 6, the method for the communication terminal B to obtain the reply prompt content of the chat initiation sentence includes:

Step S401: Acquire a topic classification to which the chat initiation sentence received by the communication terminal belongs. The method for obtaining the topic classification to which the chat initiation sentence belongs can refer to the method adopted in the simplified embodiment 1. Therefore, it is not discussed here. It is assumed that the topic acquired in this embodiment is classified as “Apple mobile phone”.

Step S402, performing data collection on user network data of the communication terminal based on the distributed cloud computing manner. Optionally, the collection of the user network data in the embodiment is implemented by using a network crawler, and the collected network data is stored by the distributed storage device, where the distributed storage device is implemented based on HDFS.

Step S403, preprocessing the user network data to obtain pre-processed text, and the pre-processing includes word segmentation processing, semantic disambiguation processing, part-of-speech tagging processing, removal of stop word processing, punctuation symbol processing, and expression character processing. Optionally, in the embodiment, when the stop word is removed, a stop word dictionary is first established, and then the words whose frequency of occurrence is high but have no practical meaning, such as "", "", "?", etc., are deleted. In the actual implementation process, the collected user network data can also be used for part-of-speech tagging and part-of-speech filtering. The part of speech filtering is based on the part-of-speech tagging. The processing of different vocabulary categories, experimental proof, adjectives and adverb pairs Clustering effect The improvement has not been much improved, so it should be removed, leaving only nouns, verbs and acronyms.

Step S404, performing text clustering on the pre-processed text by using a K-means clustering algorithm to obtain a text clustering center. In this embodiment, the K-means clustering algorithm is used to perform text clustering on the preprocessed text to obtain a text clustering center, which may include the following steps:

a. Randomly select K data as a center point, and the center point in this embodiment may be either a word or a sentence.

b. Then calculate the distance between each sample and the center point, and select the center point corresponding to the smallest distance to be the class to which it belongs. Optionally, in this embodiment, the distance between each sample and the center point is obtained by calculating the distance between the word vector corresponding to each sample and the word vector corresponding to the center point.

c. For each class, recalculate the center point (the mean of all samples in the category).

d. Repeat the iterations b and c steps until convergence, that is, the cluster center does not change.

Optionally, it is assumed that the text clustering center obtained in this embodiment is two, namely: “Apple-Mobile Phone” and “Fruit-Apple”.

Step S405, extracting a keyword of the text clustering center as a clustering topic corresponding to the text clustering center. Optionally, the clustering topic obtained in this embodiment is also two, namely “Apple-Mobile” and “Fruit-Apple”.

Step S406: Acquire a clustering topic that is closest to the topic classification to which the chat initiation sentence belongs. Optionally, when obtaining the clustering topic that is closest to the topic classification to which the chat initiation sentence belongs, the embodiment may obtain the similarity between the topic classification and the cluster topic to which the chat initiation sentence belongs, or may be pre- The relationship between the topic classification and the cluster topic is obtained. By calculating the similarity between the topic classification and the cluster topic to which the chat initiation sentence belongs, it is easy to obtain the closest topic classification ("Apple mobile phone") to which the chat initiation sentence belongs. The clustering topic is "Apple-Mobile."

Step S407: Match the chat initiation sentence in the user network data corresponding to the clustering topic to obtain a second semantic matching result. Optionally, in this embodiment, by matching the chat initiation sentence ("What is the price of the apple?") in the user network data corresponding to the clustering topic, it is easy to obtain the price corresponding to the price of the Apple mobile phone instead of The price of the apple in the fruit.

In this embodiment, the K-means clustering algorithm is used to extract the clustering topic of the user network data, and the chat initiation sentence is matched in the user network data corresponding to the clustering topic closest to the chat initiation sentence, thereby saving a large number of user networks. The matching time of the data matching the chat initiation sentence, thereby improving the speed and efficiency of the reply prompt content acquisition, and matching the chat initiation sentence by only the user network data corresponding to the cluster topic closest to the chat initiation sentence, Make the obtained reply prompt content more accurate and intelligent.

Referring to FIG. 7, an apparatus for obtaining a reply prompt content of a chat initiation sentence according to an optional embodiment of the present invention includes:

The topic database creation device 10 is configured to establish a topic database corresponding to the preset topic,

The topic classification obtaining means 20 is configured to acquire a topic classification to which the chat initiation sentence received by the communication terminal belongs;

The first semantic matching device 30 is configured to perform semantic matching on the chat initiation sentence by using a topic database corresponding to the preset topic of the topic classification, obtain a first semantic matching result, and use the first semantic matching result as a reply of the chat initiation sentence. Prompt content;

The second semantic matching device 40 is configured to collect data of the user network data of the communication terminal based on the distributed cloud computing manner if the first semantic matching result is not obtained, and perform semantic matching on the chat initiation sentence by using the user network data to obtain The second semantic matching result is used, and the second semantic matching result is used as the reply prompt content of the chat initiation sentence.

Optionally, the topic database creation device 10 includes:

a setting device configured to set a scene item associated with the preset topic and a scene option corresponding to the scene item;

The sample chat pair creation device is configured to create a sample chat pair with a preset topic as a chat topic, and the sample chat pair is used as a topic database corresponding to the preset topic, and the sample chat pair includes a sample initiation sentence, and is set according to the scene option. The sample response sentence corresponding to the sample initiation sentence.

Optionally, the topic classification obtaining device 20 includes:

Consolidate the text acquisition device, set to get the chat content of the chat initiation sentence, and chat The above chat content of the day initiation sentence and the chat initiation sentence are merged into a combined text in a text format;

a keyword extracting device configured to extract keywords of the merged text;

The topic classification determining means is configured to acquire a topic classification to which the chat initiation sentence belongs according to the keyword.

The device for obtaining the reply prompt content of the chat initiation sentence provided by the embodiment of the present invention obtains the topic classification to which the chat initiation sentence received by the communication terminal belongs, and uses the customized topic database corresponding to the preset topic with the same topic classification to chat. Initiating a sentence for semantic matching, obtaining a first semantic matching result, and collecting user network data of the communication terminal without obtaining the first semantic matching result, and using the user network data to semantically match the chat initiation sentence to obtain a second The semantic matching result solves the problem that the traditional database matching method does not necessarily obtain the chat reply prompt content matching the chat initiation sentence, thereby resulting in low intelligence of the chat and poor user experience, and fully utilizing the communication terminal. The user network data obtains the reply prompt content of the chat initiation sentence, improves the accuracy of the reply prompt content acquisition, and embodies a higher intelligent level and improves the user experience.

In the embodiment, the working process and working principle of the method for obtaining the reply prompting content of the chat initiation sentence can be referred to the working process and working principle of the method for obtaining the reply prompt content of the chat initiation sentence. The communication terminal device in the embodiment of the present invention may be a desktop computer, a tablet computer, a personal digital assistant, a mobile phone, a television, an on-board computer, a wearable communication device, or the like.

Industrial Applicability: It can be seen from the above description that the present invention obtains the topic classification of the chat initiation sentence received by the communication terminal, and uses the customized topic database corresponding to the preset topic with the same topic classification to semantically match the chat initiation sentence. Obtaining a first semantic matching result, and collecting user network data of the communication terminal without obtaining the first semantic matching result, and using the user network data to perform semantic matching on the chat initiation sentence to obtain a second semantic matching result, and solving the problem The traditional database matching method does not necessarily obtain the chat reply prompt content matching the chat initiation sentence, thereby resulting in a low degree of intelligence of the chat and a technical problem of poor user experience, and fully utilizing the user network data of the communication terminal to obtain the chat initiation. The reply prompt content of the sentence improves the accuracy of the reply prompt content acquisition, and reflects the higher level of intelligence and improvement. The user experience.

The above is only an alternative embodiment of the present invention, and is not intended to limit the present invention, and various modifications and changes can be made to the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and scope of the present invention are intended to be included within the scope of the present invention.

Claims (10)

A method for obtaining a reply prompt content of a chat initiation sentence, comprising: Establish a topic database corresponding to the preset topic; Obtaining a topic classification to which the chat initiation sentence received by the communication terminal belongs; Performing semantic matching on the chat initiation sentence by using a topic database corresponding to the preset topic of the topic classification, acquiring a first semantic matching result, and using the first semantic matching result as the chat initiation sentence Reply to the prompt content; If the first semantic matching result is not obtained, performing data collection on the user network data of the communication terminal based on the distributed cloud computing manner, and performing semantic matching on the chat initiation sentence by using the user network data to obtain the first The two semantic matching results, and the second semantic matching result is used as the reply prompt content of the chat initiation sentence. The method for obtaining a reply prompt content of a chat initiation sentence according to claim 1, wherein the establishing a topic database corresponding to the preset topic comprises: Setting a scene item associated with the preset topic, and a scene option corresponding to the scene item; Creating a sample chat pair with the preset topic as a chat topic, and using the sample chat pair as a topic database corresponding to the preset topic, the sample chat pair includes a sample initiation sentence, and setting according to the scene option A sample reply sentence corresponding to the sample initiation sentence. The method for obtaining a reply prompt content of a chat initiation sentence according to claim 2, wherein obtaining a topic classification to which the chat initiation sentence received by the communication terminal belongs includes: Obtaining the above chat content of the chat initiation sentence, and combining the chat initiation sentence and the above chat content of the chat initiation sentence into a merged text in a text format; Extracting keywords of the merged text; Obtaining a topic classification to which the chat initiation sentence belongs according to the keyword. The method for obtaining a reply prompt content of a chat initiation sentence according to claim 3, wherein the chat initiation sentence is semantically matched by using a topic database corresponding to the preset topic of the topic classification, and obtaining the first Semantic matching results include: Obtaining a topic database corresponding to the preset topic that is the same as the topic classification; Collecting content information of a scene item associated with the topic classification to obtain scene information; And matching, in the topic database, a sample initiation sentence that is the same as the chat initiation sentence, and acquiring a first semantic matching result according to the scenario information. The method for obtaining the reply prompt content of the chat initiation sentence according to any one of claims 1-4, wherein the semantic initiation of the chat initiation sentence is performed by using the user network data, and obtaining the second semantic matching result comprises: Pre-processing the user network data to obtain pre-processed text, the pre-processing including word segmentation processing, semantic disambiguation processing, part-of-speech tagging processing, removing stop word processing, punctuation symbol processing, and emoticon character processing; The K-means clustering algorithm is used to perform text clustering on the pre-processed text to obtain a text clustering center; Extracting a keyword of the text clustering center as a clustering topic corresponding to the text clustering center; Obtaining the clustering topic that is closest to the topic classification to which the chat initiation sentence belongs; Matching the chat initiation sentence in the user network data corresponding to the clustering topic to obtain a second semantic matching result. The method for obtaining a reply prompt content of a chat initiation sentence according to claim 5, wherein the scene entry comprises: Relationship entry, name entry, gender entry, age entry, instant messaging account entry, email address entry, home address entry, occupation category entry, job entry, work unit entry, unit of the communication terminal that sends and receives the chat initiation sentence Address entry, Bank account entry, friend impression entry, hobbyist entry, circle of friends status entry, mood entry, recent topic of interest entry, current communication status entry, scene image entry, time entry, holiday entry, season entry, geographic location information entry, distance entry a communication frequency entry, a communication frequency entry, a communication duration entry, and a selection mode entry for initiating historical communication, wherein the selection manner includes starting a communication mode from the address book, initiating a communication mode from the historical call record, and initiating a communication mode from the short message communication module. Start the communication method from the dial pad. The method for obtaining the reply prompt content of the chat initiation sentence according to claim 6, wherein the collecting the content information of the scene image entry of the communication terminal that sends or receives the chat initiation sentence in the scene entry comprises: Collecting a scene image of the communication terminal that sends or receives the chat initiation sentence; Extracting a region of interest of the scene training image by using a Gaussian difference DOG operator, and calculating a scale-invariant feature transform SIFT feature of the region of interest of the scene training image; The K-means clustering algorithm is used to cluster the SIFT features of the region of interest of the scene training image to obtain a plurality of cluster centers, and a visual word dictionary composed of visual words corresponding to each of the cluster centers is established. ; Extracting a region of interest of the scene image using a DOG operator, and matching a visual word in the visual word dictionary that is closest to a SIFT feature of the region of interest of the scene image; And classifying the scene image by using a pre-trained support vector machine classifier according to the distribution of the visual words of the region of interest of the scene image, and obtaining the content of the scene image item of the communication terminal that sends or receives the chat initiation sentence. information. A device for obtaining a reply prompt content of a chat initiation sentence, comprising: a topic database creation device, configured to establish a topic database corresponding to the preset topic; The topic classification obtaining device is configured to obtain a topic classification to which the chat initiation sentence received by the communication terminal belongs; a first semantic matching device, configured to perform semantic matching on the chat initiation sentence by using a topic database corresponding to the preset topic of the topic classification, acquire a first semantic matching result, and match the first semantic The result is the reply prompt content of the chat initiation sentence; a second semantic matching device, configured to collect data of user network data of the communication terminal based on a distributed cloud computing manner, and use the user network data to perform the chat if the first semantic matching result is not obtained The initiating sentence performs semantic matching, obtains a second semantic matching result, and uses the second semantic matching result as the reply prompt content of the chat initiation sentence. The apparatus for obtaining a reply prompt content of a chat initiation sentence according to claim 8, wherein the topic database creation means comprises: a setting device configured to set a scene item associated with the preset topic, and a scene option corresponding to the scene item; a sample chat pair creation device, configured to create a sample chat pair with the preset topic as a chat topic, and use the sample chat pair as a topic database corresponding to the preset topic, where the sample chat pair includes a sample initiation a sentence, a sample reply sentence corresponding to the sample initiation sentence set according to the scene option. The apparatus for obtaining a reply prompt content of a chat initiation sentence according to claim 9, wherein the topic classification acquisition means comprises: a merged text obtaining device, configured to acquire the above chat content of the chat initiation sentence, and merge the chat initiation sentence and the chat content of the chat initiation sentence into a merged text in a text format; a keyword extracting device configured to extract keywords of the merged text; The topic classification determining means is configured to acquire a topic classification to which the chat initiation sentence belongs according to the keyword.

Images