WO2021051558A1 - Knowledge graph-based question and answer method and apparatus, and storage medium - Google Patents

Abstract

A knowledge graph-based question and answer method and apparatus, and a storage medium, the method comprising: acquiring a question and answer sentence inputted by a user, performing word segmentation on the question and answer sentence and, by means of a trained entity element recognition model and label recognition model, respectively acquiring entity elements in the question and answer sentence and labels corresponding to the entity elements (S10); inputting the entity elements, the labels, and the question and answer sentence into a Bayesian classifier, calculating the degree of matching between preset templates in the Bayesian classifier and the question and answer sentence, and determining the preset template with the highest degree of matching as a query template (S20); inputting the entity elements and the labels into the query template to obtain a corresponding query sentence, and inputting the query sentence into a knowledge graph for querying to obtain a corresponding question and answer result (S30).

Description

Question answering method, device and storage medium based on knowledge graph To

This application claims the priority of a Chinese patent application filed with the Chinese Patent Office on September 18, 2019, the application number is 201910885936.X, and the invention title is "Question and Answer Method, Apparatus, and Storage Medium Based on Knowledge Graph", and the entire content of which is approved The reference is incorporated in the application.

Technical field

This application relates to the field of information processing technology, and in particular to a question and answer method, device and storage medium based on a knowledge graph.

Background technique

With the development of Internet technology and the emergence of large-scale network data resources, people hope to obtain valuable information accurately and quickly from massive amounts of Internet data, which promotes the widespread application of search-type question and answer systems. Question answering system is an advanced form of information retrieval system. It can use accurate and concise natural language to answer users' questions in natural language. In the current field of artificial intelligence and natural language processing, the development and improvement of question answering systems is also a research direction that has attracted much attention and has broad development prospects.

At present, the traditional question answering system is based on a certain question and answer corpus training model, the user's natural language is processed and input into the trained model, and the result is obtained by querying similar corpus in the model. However, the accuracy of this question answering system depends on the coverage of the training corpus. When the question input by the user is more complicated, the question answering result output by the traditional question answering system is not accurate.

Summary of the invention

The main purpose of this application is to provide a question and answer method, device and storage medium based on a knowledge graph, aiming to solve the technical problem of inaccurate question and answer results output by a traditional question and answer system.

In order to achieve the above purpose, this application provides a question and answer method based on a knowledge graph, which includes the following steps:

Obtain training corpus through web crawler technology, and perform word segmentation on the training corpus;

Receiving the artificial entity element and artificial labeling of the word obtained after word segmentation, and the artificial entity element and artificial label corresponding to the word are obtained;

Inputting the artificial entity element and training corpus into a preset entity element recognition model to train the entity element recognition model;

Input the artificial label and training corpus into a preset label recognition model to train the label recognition model;

Acquiring the question and answer sentence input by the user, segmenting the question and answer sentence, and obtaining the entity element in the question and answer sentence and the label corresponding to the entity element through the entity element recognition model and the label recognition model completed through training;

The entity element, the label, and the question and answer sentence are input to the Bayesian classifier, and the matching degree between each preset template in the Bayesian classifier and the question and answer sentence is calculated, and the one with the highest matching degree is calculated The preset template is determined to be the query template;

The entity element and the label are input into the query template to obtain the corresponding query sentence, and the query sentence is input into the knowledge graph for query, and the corresponding question and answer result is obtained.

The present application also provides a device that includes a memory, a processor, and computer-readable instructions stored on the memory and capable of running on the processor, and the computer-readable instructions are used by the processor When executed, the steps of the question answering method based on the knowledge graph as described above are realized.

The present application also provides a non-volatile computer-readable storage medium having computer-readable instructions stored on the non-volatile computer-readable storage medium, and when the computer-readable instructions are executed by a processor, the aforementioned Steps of question answering method based on knowledge graph.

The application discloses a question and answer method, device and storage medium based on a knowledge graph. The method first obtains the question and answer sentence input by the user, performs word segmentation on the question and answer sentence, and completes the entity element recognition model and label recognition through training The model obtains the entity elements in the question and answer sentence and the label corresponding to the entity element respectively; inputs the entity element, the label, and the question and answer sentence into the Bayesian classifier, and calculates the Bayesian classifier The degree of matching between each preset template and the question and answer sentence, and the preset template with the highest matching degree is determined as the query template; the entity element and the label are input into the query template to obtain the corresponding query sentence, The query sentence is input into the knowledge graph for query, and the corresponding question and answer result is obtained. This application analyzes the user’s question and answer sentences of the trained entity element recognition model and the label recognition model, and obtains the entity elements and labels of the question and answer sentence by training the entity element recognition model and the label recognition model. Hierarchical mining; determine the most suitable query template, and generate the corresponding query sentence, get the corresponding question and answer result according to the knowledge map, the whole process reduces the dependence on the training corpus, avoids the question and answer system missed detection and error detection, thereby improving the question and answer The accuracy of the Q&A results output by the system.

Description of the drawings

FIG. 1 is a schematic diagram of the device structure of the hardware operating environment involved in the solution of the embodiment of the present application;

2 is a schematic flowchart of an embodiment of a question answering method based on a knowledge graph of this application;

FIG. 3 is a schematic flowchart of another embodiment of the question answering method based on the knowledge graph of this application;

FIG. 4 is a schematic flowchart of another embodiment of the question answering method based on the knowledge graph of this application.

The realization, functional characteristics, and advantages of the purpose of this application will be further described in conjunction with the embodiments and with reference to the accompanying drawings.

detailed description

In order to make the purpose, technical solutions, and advantages of this application clearer, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the optional embodiments described here are only used to explain the present application, and are not used to limit the present application.

As shown in FIG. 1, FIG. 1 is a schematic diagram of a terminal structure of a hardware operating environment involved in a solution of an embodiment of the present application.

The terminal of this application is a device, and the device may be a terminal device with a storage function such as a mobile phone, a computer, or a mobile computer.

As shown in FIG. 1, the terminal may include: a processor 1001, such as a CPU, a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. Among them, the communication bus 1002 is used to implement connection and communication between these components. The user interface 1003 may include a display screen (Display) and an input unit such as a keyboard (Keyboard), and the optional user interface 1003 may also include a standard wired interface and a wireless interface. The network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface). The memory 1005 can be a high-speed RAM memory or a stable memory (non-volatile memory), such as disk storage. Optionally, the memory 1005 may also be a storage device independent of the aforementioned processor 1001.

Optionally, the terminal may also include a camera, a Wi-Fi module, etc., which will not be repeated here.

Those skilled in the art can understand that the terminal structure shown in FIG. 1 does not constitute a limitation on the terminal, and may include more or fewer components than shown in the figure, or combine some components, or arrange different components.

Optionally, the terminal may also include a camera, a Wi-Fi module, etc., which will not be repeated here.

Those skilled in the art can understand that the terminal structure shown in FIG. 1 does not constitute a limitation on the terminal, and may include more or fewer components than shown in the figure, or combine some components, or arrange different components.

In the terminal shown in FIG. 1, the network interface 1004 is mainly used to connect to a back-end server and communicate with the back-end server; the user interface 1003 mainly includes an input unit such as a keyboard. The keyboard includes a wireless keyboard and a wired keyboard for connecting to a client. Perform data communication with the client; and the processor 1001 can be used to call the computer readable instructions stored in the memory 1005 and execute the steps of the question and answer method based on the knowledge graph.

The optional embodiments of the device are basically the same as the following embodiments of the question and answer method based on the knowledge graph, and will not be repeated here.

Please refer to FIG. 1. FIG. 1 is a schematic diagram of the device structure of the hardware operating environment involved in the solution of the embodiment of the present application. The question answering method based on the knowledge graph provided in this embodiment includes the following steps:

Step S10: Obtain the question and answer sentence input by the user, perform word segmentation on the question and answer sentence, and obtain the entity element in the question and answer sentence and the label corresponding to the entity element through the entity element recognition model and the label recognition model completed by training. ；

To obtain the question and answer sentences input by the user, it should be understood that the question and answer sentences expressed by the user can be obtained by means of voice recognition, and the question and answer sentences input by the user can also be obtained by other methods, which is not limited in this embodiment. Based on NLP technology, the obtained question and answer sentences are segmented, and the segmented question and answer sentences are passed through the entity element recognition model to extract the entity elements of the question and answer sentences; the question and answer sentences after the word segmentation are passed through the label recognition model to extract the question and answer sentences. label. In order to elaborate on the relationship between question and answer sentences, entity elements and labels, take the question and answer sentence "What is the relationship between Huang Xiaoming and Changsheng Medicine" as an example. In the above question and answer sentence, "Huang Xiaoming" is the entity element, and the corresponding label is person, "Changsheng "Medicine" is an entity element, and the corresponding label is a company.

Step S20: Input the entity element, the label, and the question and answer sentence into a Bayesian classifier, calculate the matching degree between each preset template in the Bayesian classifier and the question and answer sentence, and compare the matching The preset template with the highest degree is determined as the query template;

Input the question and answer sentences input by the user, the entity elements and labels corresponding to the question and answer sentences into the preset Bayesian classifier. The Bayesian classifier uses the prior probability of an object to calculate the object using the Bayesian formula For the probability of belonging to a certain class, the class with the largest posterior probability is selected as the class to which the object belongs. Optionally, there are three preset templates in the Bayesian classifier, namely, relational query template, entity query template, and attribute query template, which are combined by calculating the matching degree of each preset template with entity elements, tags, and question and answer sentences The contextual semantics of the question and answer sentence determines the template with the highest matching degree as the query template.

Step S30, input the entity element and the label into the query template to obtain the corresponding query sentence, and input the query sentence into the knowledge graph for query, and obtain the corresponding question and answer result.

After the query template is obtained, the entity elements and tags corresponding to the question and answer statement are input into the query template to obtain the corresponding query statement. In addition, a knowledge graph is also preset in this embodiment, and the query sentence is input into the knowledge graph, and the corresponding question and answer result is obtained by means of vector feature matching.

This embodiment first obtains the question and answer sentence input by the user, performs word segmentation on the question and answer sentence, and obtains the entity element in the question and answer sentence and the corresponding entity element through the entity element recognition model and the label recognition model completed by training. Label; input the entity element, the label, and the question and answer sentence into the Bayesian classifier, calculate the matching degree between each preset template in the Bayesian classifier and the question and answer sentence, and compare the matching degree The highest preset template is determined to be the query template; the entity elements and the tags are input into the query template to obtain the corresponding query sentence, and the query sentence is input into the knowledge graph for query, and the corresponding Q&A results. Through the analysis of the user’s question and answer sentences of the trained entity element recognition model and the label recognition model, the entity elements and labels of the question and answer sentences are obtained by training the entity element recognition model and the label recognition model, and the user’s question and answer sentences are deeply analyzed. Mining; determine the most suitable query template, and generate the corresponding query sentence, and get the corresponding question and answer result according to the knowledge graph. The whole process reduces the dependence on the training corpus, avoids the question and answer system's missed detection and error detection, thereby improving the output of the question and answer system The accuracy of the Q&A results.

Further, please refer to FIG. 3, which is a schematic flowchart of another embodiment of the question answering method based on the knowledge graph of this application. Step S10 obtains the question and answer sentence input by the user, and before performing word segmentation on the question and answer sentence, further includes:

Step S40: Obtain training corpus through web crawler technology, and perform word segmentation on the training corpus;

Before all the steps, it is also necessary to use the existing data to train the entity element recognition model and the label recognition model. First, the web crawler technology is used to obtain a large amount of information from the existing database as training corpus, and the obtained training corpus is segmented. Among them, the web crawler technology is a program that automatically captures information on the World Wide Web according to certain rules or Scripts can automatically update a method of information acquisition and information retrieval of the stored content. For example, for the question and answer sentence "What is the relationship between Huang Xiaoming and Changsheng Medicine", crawlers can be crawled from related databases such as the national enterprise information publicity system, news database, and enterprise credit database to obtain relevant training corpus.

Step S50, receiving the artificial entity element and artificial label of the word obtained after word segmentation, and the artificial entity element and artificial label corresponding to the word are received;

After the training corpus is segmented, the obtained words are manually labeled with entity elements and labels. Special attention is paid to the fact that if the entity elements include multiple Chinese characters, the labels should be manually labeled at the preset position of each Chinese character; and As an implementation manner, the type of the label may be determined according to the type of the preset query template.

For example, the preset query templates include relationship query templates, entity query templates, and attribute query templates. The entity elements are "Huang Xiaoming" and "Changsheng Medicine", and the result of manual labeling is "Huang person Xiao person Ming person "And "long company health company medical company medicine company".

Step S60, input the artificial entity element and training corpus into a preset entity element recognition model to train the entity element recognition model;

For the training of the entity element recognition model, after the manual labeling of the entity elements is completed, the artificial entity elements and training corpus are input into the preset entity element recognition model, and the preset entity element recognition model is trained.

Step S70: Input the artificial label and training corpus into a preset label recognition model to train the label recognition model.

For the training of the label recognition model, after the manual labeling is completed, the artificial label and training corpus are input into the preset label recognition model, and the preset label recognition model is trained.

This embodiment uses crawler technology to obtain a large number of training corpora, so as to provide the model with data required for training, and uses manual standard methods to perform entity element labeling and label labeling on the training corpus to train the entity element recognition model and the label recognition model, thereby ensuring The accuracy of extracting entity elements by the entity element recognition model and the accuracy of extracting tags by the label recognition model.

Further, the step of training the entity element recognition model includes:

Step S61, extracting the training corpus through the entity element recognition model to obtain corresponding extracted entity elements;

In this embodiment, the training corpus is input into the entity element recognition model, the training corpus is extracted through the entity element recognition model, and the entity elements extracted by the entity element recognition model are used as the extracted entity elements. It should be understood that since the accuracy of the entity element recognition model has not been judged, the entity element recognition model may extract wrong entity elements, that is, entity elements that do not belong to the training corpus.

Step S62: Determine the entity element that overlaps the extracted entity element among the artificial entity elements as an entity element set;

In the above steps, the training corpus is manually labeled with entity elements, the artificial entity elements are compared with the extracted entity elements, and the entity elements that overlap with the extracted entity elements among the artificial entity elements are taken as the entity element set. It is easy to understand that the entity elements in the entity element set must be the correct entity elements, that is, they must belong to the input training corpus.

Step S63, calculating the accuracy of the entity element recognition model according to the entity element set, the artificial entity element and the extracted entity element;

After the entity element set is obtained, the preset formula is used to calculate the entity element set, artificial entity element and extracted entity element to obtain the accuracy of the trained entity element recognition model.

In step S64, the entity element recognition model whose accuracy exceeds the preset first accuracy is used as the entity element recognition model completed by the training.

In this embodiment, if the accuracy of the obtained entity element recognition model does not exceed the preset first accuracy, it means that the entity element recognition model is not accurate enough in extracting the entity elements, and the wrong entity will be extracted, and the training corpus will continue to be used as Input and train the entity element recognition model until the accuracy of the entity element recognition model exceeds the preset first accuracy.

This embodiment uses training corpus and artificial entity elements to train the entity element recognition model, and by calculating the accuracy of the entity element recognition model to ensure that the trained entity element recognition model meets the preset accuracy, thereby ensuring that the entity element recognition model can Accurately extract the entity elements in question and answer sentences.

Further, the step of calculating the accuracy of the entity element recognition model based on the entity element set, the artificial entity element and the extracted entity element includes:

Step S631: Divide the set of entity elements by the extracted entity elements to obtain the accuracy rate of the entity element recognition model;

In this embodiment, since the entity elements in the entity element set must be the correct entity elements, and the entity elements extracted by the entity element recognition model may not belong to the entity elements of the training corpus; therefore, the value of the entity elements in the entity element set Divide the value of the entity element extracted by the entity element recognition model, and use the calculation result as the accuracy rate of the entity element recognition model.

Step S632: Divide the set of entity elements by the artificial entity elements to obtain the recall rate of the entity element recognition model;

In this embodiment, the value of the entity element in the entity element set is divided by the value of the entity element manually labeled, and the calculation result is used as the recall rate of the entity element recognition model.

Step S633: Calculate the product of the accuracy rate and the recall rate and the sum of the accuracy rate and the recall rate, divide the product by the sum value, and multiply the calculated result by the preset value as the entity element recognition model The first F value;

After obtaining the recall rate and accuracy rate of the entity element recognition model, multiply the accuracy rate and the recall rate, add the accuracy rate and the recall rate, and divide the product of the accuracy rate and the recall rate. Multiply the obtained calculation result by a preset value as the first F value of the entity element recognition model. Optionally, the preset value is 2, that is, the first F value = 2×(accuracy rate×recall rate)/( Accuracy rate + recall rate).

Step S634: According to the accuracy rate, the recall rate and the first F value of the entity element recognition model, the accuracy of the entity element recognition model is calculated.

After obtaining the accuracy rate, recall rate and the first F value, the accuracy of the entity element recognition model can be obtained according to the above three elements. By setting different weights for the accuracy rate, recall rate and the first F value of the entity element recognition model To get the accuracy, it is easy to understand that the design of the weight ratio corresponding to each value can be drawn up by the developer.

In this embodiment, the accuracy, recall, and first F value of the entity element recognition model are calculated according to the entity element set and the extracted entity elements, and the entity element recognition model is obtained by accurate calculation based on the accuracy, recall rate and the first F value. Accuracy.

Further, the step of training the label recognition model includes:

Step S71: Extract the above-mentioned training corpus through the label recognition model to obtain the corresponding extracted label;

In this embodiment, the training corpus is input into the label recognition model, the training corpus is extracted through the label recognition model, and the label extracted by the label recognition model is used as the extracted label corresponding to the training corpus. It should be understood that, since the accuracy of the label recognition model has not been judged, the label recognition model may extract wrong labels, that is, labels that are not part of the training corpus.

Step S72: Determine the label that overlaps the extracted label among the artificial labels as a label set;

Since the training corpus is manually labelled in the above steps, the artificial label is compared with the extracted label, and the label in the artificial label that is coincident with the extracted label is used as the label set. It is easy to understand that the tags in the tag set must be the correct tags, that is, the tags must belong to the input training corpus.

Step S73: Calculate the accuracy of the label recognition model according to the label set, the artificial label and the extracted label;

After the label set is obtained, a preset formula is used to perform calculations based on the label set, manual labels, and extracted labels to obtain the accuracy of the trained label recognition model.

In step S74, the label recognition model whose accuracy exceeds the preset second accuracy is used as the trained label recognition model.

In this embodiment, if the accuracy of the obtained label recognition model does not exceed the preset second accuracy, it means that the label recognition model is not accurate enough for label extraction, and the wrong label will be extracted. Then the training corpus will continue to be used as the input training place. The label recognition model is described until the accuracy of the label recognition model exceeds the preset second accuracy.

This embodiment uses training corpus and artificial tags to train the label recognition model, and by calculating the accuracy of the label recognition model to ensure that the trained label recognition model meets the preset accuracy, thereby ensuring that the label recognition model can accurately extract question and answer sentences In the label.

Further, the step of calculating the accuracy of the label recognition model according to the label set, the artificial label and the extracted label includes:

Step S731: Divide the label set by the extracted label to obtain the accuracy rate of the label recognition model;

In this embodiment, since the tags in the tag set must be correct tags, and the tags extracted by the tag recognition model may not belong to the tags of the training corpus; therefore, the value of the tags in the tag set is divided by the value extracted by the tag recognition model The numerical value of the label, and the calculation result is used as the accuracy rate of the label recognition model.

Step S732: Divide the label set by the artificial label to obtain the recall rate of the label recognition model;

In this embodiment, the numerical value of the label in the label set is divided by the numerical value of the manually labeled label, and the calculation result is used as the recall rate of the label recognition model.

Step S733: Calculate the product of the accuracy rate and the recall rate and the sum of the accuracy rate and the recall rate, divide the product by the sum value, and multiply the obtained calculation result by a preset value as the label recognition model Second F value

After the recall rate and accuracy rate of the label recognition model are obtained, the accuracy rate is multiplied by the recall rate, the accuracy rate and the recall rate are added, and the product of the accuracy rate and the recall rate is divided. Multiply the obtained calculation result by a preset value as the second F value of the label recognition model. Optionally, the preset value is 2, that is, the second F value = 2×(accuracy rate×recall rate)/(accurate Rate + recall rate).

Step S734: According to the accuracy rate, the recall rate and the second F value of the tag recognition model, the accuracy of the tag recognition model is calculated.

After obtaining the accuracy rate, recall rate and the second F value, the accuracy of the label recognition model can be obtained based on the above three elements, and the accuracy can be calculated by setting different weights for the accuracy rate, recall rate and the second F value of the label recognition model It is easy to understand that the design of the weight ratio corresponding to each value can be drawn up by the developer.

In this embodiment, the accuracy rate, recall rate, and second F value of the label recognition model are calculated according to the label set and extracted tags, and the accuracy of the label recognition model is obtained by accurate calculation according to the accuracy rate, recall rate and the second F value.

Further, please refer to FIG. 4, which is a schematic flowchart of another embodiment of the question answering method based on the knowledge graph of this application. After the step S50 obtains the artificial entity elements and the artificial tags corresponding to the words, the method further includes:

Step S80, using artificial entity elements and artificial tags as inputs of a preset TransE algorithm, so that the artificial entity elements and the artificial tags are embedded in a low-dimensional vector space to generate a corresponding vector template;

The TransE algorithm is also preset in this embodiment, and the TransE algorithm is a distributed vector representation based on entities and relationships. The artificial entity elements and artificial tags are input into the preset TransE algorithm, and the artificial entity elements and artificial tags are embedded in the low-dimensional vector space through the TransE algorithm algorithm to generate the corresponding vector template.

In step S90, the vector template is stored in the graph database to construct a corresponding knowledge graph.

After the vector template is obtained, the vector template is stored in the graph database, and the corresponding knowledge graph is constructed according to the multiple vector templates stored in the graph database. The construction method of the knowledge graph is not described in this embodiment.

This embodiment uses a preset TransE algorithm to obtain a vector template according to artificial entity elements and artificial tags, and constructs a corresponding knowledge map according to the vector template, so as to ensure the comprehensiveness and accuracy of the data of the knowledge map.

Further, the step of inputting the query sentence into the knowledge graph for query, and obtaining the corresponding question and answer result includes:

Step S31, vectorizing the query sentence to generate a corresponding vector set;

In this embodiment, after the query sentence is obtained, the query sentence is vectorized by the NLP algorithm to generate the corresponding vector set. It should be understood that the vectorization of the query sentence can also be realized in other ways, and this embodiment will not do it here. limit.

Step S32, matching the vector set with the vector template in the knowledge graph to obtain a corresponding question and answer result.

After the vector set is obtained, the vector set is matched with the vector template in the knowledge graph. Optionally, the vector that best matches the vector set in the knowledge graph is determined by calculating the matching degree between the vectors. Template, and then parse the vector template to obtain the corresponding question and answer result.

In this embodiment, the query sentence is vectorized, and the result of the question and answer in the knowledge graph is determined more accurately through the matching of the vector set and the vector template, thereby ensuring the accuracy of the question and answer result.

In addition, the embodiment of the present application also proposes a non-volatile computer-readable storage medium, the non-volatile computer-readable storage medium stores computer-readable instructions, and when the computer-readable instructions are executed by a processor The operation of the question answering method based on the knowledge graph as described above is realized.

The optional embodiments of the non-volatile computer-readable storage medium of the present application are basically the same as the above-mentioned embodiments of the question and answer method based on the knowledge graph, and will not be repeated here.

It should be noted that in this article, the terms "include", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or system including a series of elements not only includes those elements, It also includes other elements that are not explicitly listed, or elements inherent to the process, method, article, or system. Without more restrictions, the element defined by the sentence "including a..." does not exclude the existence of other identical elements in the process, method, article, or system that includes the element.

The serial numbers of the foregoing embodiments of the present application are only for description, and do not represent the advantages and disadvantages of the embodiments.

Through the description of the above embodiments, those skilled in the art can clearly understand the above The method of the embodiment can be realized by means of software plus the necessary general hardware platform, of course, it can also be realized through Over hardware, but in many cases the former is a better implementation. Based on this understanding, the technical solution of this application essentially or the part that contributes to the existing technology can be embodied in the form of a software product. Now, the computer software product is stored in a storage medium (such as ROM/RAM, floppy disk, optical disk) The disk) includes several instructions to enable a terminal device (which may be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to execute the method described in each embodiment of the present application.

The above are only optional embodiments of this application, and do not limit the scope of this application. Any equivalent structure or equivalent process transformation made using the content of the description and drawings of this application, or directly or indirectly applied to other related technologies In the same way, all fields are included in the scope of patent protection of this application.

Claims (20)

A question and answer method based on knowledge graph, which includes the following steps: Obtain training corpus through web crawler technology, and perform word segmentation on the training corpus; Receiving the artificial entity element and artificial labeling of the word obtained after word segmentation, and the artificial entity element and artificial label corresponding to the word are obtained; Inputting the artificial entity element and training corpus into a preset entity element recognition model to train the entity element recognition model; Input the artificial label and training corpus into a preset label recognition model to train the label recognition model; Acquiring the question and answer sentence input by the user, segmenting the question and answer sentence, and obtaining the entity element in the question and answer sentence and the label corresponding to the entity element through the entity element recognition model and the label recognition model completed through training; The entity element, the label, and the question and answer sentence are input to the Bayesian classifier, and the matching degree between each preset template in the Bayesian classifier and the question and answer sentence is calculated, and the one with the highest matching degree The preset template is determined to be the query template; The entity element and the label are input into the query template to obtain the corresponding query sentence, and the query sentence is input into the knowledge graph for query, and the corresponding question and answer result is obtained. The question answering method based on the knowledge graph of claim 1, wherein the step of training the entity element recognition model comprises: Extracting the training corpus through the entity element recognition model to obtain corresponding extracted entity elements; Determining entity elements that overlap with the extracted entity elements among the artificial entity elements as an entity element set; Calculating the accuracy of the entity element recognition model according to the entity element set, the artificial entity element, and the extracted entity element; The entity element recognition model whose accuracy exceeds the preset first accuracy is used as the entity element recognition model completed by training. The question and answer method based on the knowledge graph of claim 2, wherein the step of calculating the accuracy of the entity element recognition model based on the entity element set, the artificial entity element, and the extracted entity element include: Dividing the set of entity elements by the extracted entity elements to obtain the accuracy rate of the entity element recognition model; Dividing the set of entity elements by the artificial entity elements to obtain the recall rate of the entity element recognition model; Calculate the product of the accuracy rate and the recall rate and the sum of the accuracy rate and the recall rate, divide the product by the sum value, and multiply the calculated result by the preset value as the first of the entity element recognition model F value; According to the accuracy rate, the recall rate and the first F value of the entity element recognition model, the accuracy of the entity element recognition model is calculated. The question answering method based on the knowledge graph of claim 1, wherein the step of training the label recognition model comprises: Extracting the above-mentioned training corpus through the label recognition model to obtain the corresponding extracted label; Determining that the tags in the manual tags that overlap with the extracted tags are used as a tag set; Calculating the accuracy of the label recognition model according to the label set, the artificial label, and the extracted label; The label recognition model whose accuracy exceeds the preset second accuracy is used as the trained label recognition model. The question answering method based on the knowledge graph of claim 4, wherein the step of calculating the accuracy of the label recognition model according to the label set, the artificial label and the extracted label comprises: Dividing the label set by the extracted label to obtain the accuracy rate of the label recognition model; Dividing the label set by the artificial label to obtain the recall rate of the label recognition model; Calculate the product of the accuracy rate and the recall rate and the sum of the accuracy rate and the recall rate, divide the product by the sum value, and multiply the calculated result by the preset value as the second F of the label recognition model value; According to the accuracy rate, the recall rate and the second F value of the tag recognition model, the accuracy of the tag recognition model is calculated. The question answering method based on the knowledge graph of claim 1, wherein after the step of obtaining the artificial entity elements and artificial tags corresponding to the words, the method comprises: Using artificial entity elements and artificial tags as inputs of a preset TransE algorithm, so that the artificial entity elements and the artificial tags are embedded in a low-dimensional vector space to generate a corresponding vector template; The vector template is stored in the input graph database to construct a corresponding knowledge graph. The question answering method based on the knowledge graph of claim 6, wherein the step of inputting the query sentence into the knowledge graph for query, and obtaining the corresponding question and answer result comprises: Vectorizing the query sentence to generate a corresponding vector set; The vector set is matched with the vector template in the knowledge graph to obtain a corresponding question and answer result. A device, wherein the device includes: a memory, a processor, and computer-readable instructions stored on the memory and capable of running on the processor, and when the computer-readable instructions are executed by the processor , Perform the following steps: Obtain training corpus through web crawler technology, and perform word segmentation on the training corpus; Receiving the artificial entity element and artificial labeling of the word obtained after word segmentation, and the artificial entity element and artificial label corresponding to the word are obtained; Inputting the artificial entity element and training corpus into a preset entity element recognition model to train the entity element recognition model; Input the artificial label and training corpus into a preset label recognition model to train the label recognition model; Acquiring the question and answer sentence input by the user, segmenting the question and answer sentence, and obtaining the entity element in the question and answer sentence and the label corresponding to the entity element through the entity element recognition model and the label recognition model completed through training; The entity element, the label, and the question and answer sentence are input to the Bayesian classifier, and the matching degree between each preset template in the Bayesian classifier and the question and answer sentence is calculated, and the one with the highest matching degree The preset template is determined to be the query template; The entity element and the label are input into the query template to obtain the corresponding query sentence, and the query sentence is input into the knowledge graph for query, and the corresponding question and answer result is obtained. 8. The apparatus of claim 8, when the computer-readable instructions are executed by the processor, the following steps are further performed: Extracting the training corpus through the entity element recognition model to obtain corresponding extracted entity elements; Determining entity elements that overlap with the extracted entity elements among the artificial entity elements as an entity element set; Calculating the accuracy of the entity element recognition model according to the entity element set, the artificial entity element, and the extracted entity element; The entity element recognition model whose accuracy exceeds the preset first accuracy is used as the entity element recognition model completed by training. 9. The apparatus of claim 9, when the computer-readable instructions are executed by the processor, the following steps are further performed: Dividing the set of entity elements by the extracted entity elements to obtain the accuracy rate of the entity element recognition model; Dividing the set of entity elements by the artificial entity elements to obtain the recall rate of the entity element recognition model; Calculate the product of the accuracy rate and the recall rate and the sum of the accuracy rate and the recall rate, divide the product by the sum value, and multiply the calculated result by the preset value as the first of the entity element recognition model F value; According to the accuracy rate, the recall rate and the first F value of the entity element recognition model, the accuracy of the entity element recognition model is calculated. 8. The apparatus of claim 8, when the computer-readable instructions are executed by the processor, the following steps are further performed: Extracting the above-mentioned training corpus through the label recognition model to obtain the corresponding extracted label; Determining that the tags in the manual tags that overlap with the extracted tags are used as a tag set; Calculating the accuracy of the label recognition model according to the label set, the artificial label, and the extracted label; The label recognition model whose accuracy exceeds the preset second accuracy is used as the trained label recognition model. The apparatus of claim 11, when the computer-readable instructions are executed by the processor, the following steps are further performed: Dividing the label set by the extracted label to obtain the accuracy rate of the label recognition model; Dividing the label set by the artificial label to obtain the recall rate of the label recognition model; Calculate the product of the accuracy rate and the recall rate and the sum of the accuracy rate and the recall rate, divide the product by the sum value, and multiply the calculated result by the preset value as the second F of the label recognition model value; According to the accuracy rate, the recall rate and the second F value of the tag recognition model, the accuracy of the tag recognition model is calculated. 8. The apparatus of claim 8, when the computer-readable instructions are executed by the processor, the following steps are further performed: Using artificial entity elements and artificial tags as inputs of a preset TransE algorithm, so that the artificial entity elements and the artificial tags are embedded in a low-dimensional vector space to generate a corresponding vector template; The vector template is stored in the input graph database to construct a corresponding knowledge graph. The apparatus according to claim 13, when the computer-readable instructions are executed by the processor, the following steps are further executed: Vectorizing the query sentence to generate a corresponding vector set; The vector set is matched with the vector template in the knowledge graph to obtain a corresponding question and answer result. A non-volatile computer-readable storage medium, wherein computer-readable instructions are stored on the non-volatile computer-readable storage medium, and when the computer-readable instructions are executed by a processor, the following steps are performed: Obtain training corpus through web crawler technology, and perform word segmentation on the training corpus; Receiving the artificial entity element and artificial labeling of the word obtained after word segmentation, and the artificial entity element and artificial label corresponding to the word are obtained; Inputting the artificial entity element and training corpus into a preset entity element recognition model to train the entity element recognition model; Input the artificial label and training corpus into a preset label recognition model to train the label recognition model; Acquiring the question and answer sentence input by the user, segmenting the question and answer sentence, and obtaining the entity element in the question and answer sentence and the label corresponding to the entity element through the entity element recognition model and the label recognition model completed through training; The entity element, the label, and the question and answer sentence are input to the Bayesian classifier, and the matching degree between each preset template in the Bayesian classifier and the question and answer sentence is calculated, and the one with the highest matching degree The preset template is determined to be the query template; The entity element and the label are input into the query template to obtain the corresponding query sentence, and the query sentence is input into the knowledge graph for query, and the corresponding question and answer result is obtained. The non-volatile computer-readable storage medium according to claim 15, when the computer-readable instructions are executed by the processor, the following steps are further executed: Extracting the training corpus through the entity element recognition model to obtain corresponding extracted entity elements; Determining entity elements that overlap with the extracted entity elements among the artificial entity elements as an entity element set; Calculating the accuracy of the entity element recognition model according to the entity element set, the artificial entity element, and the extracted entity element; The entity element recognition model whose accuracy exceeds the preset first accuracy is used as the entity element recognition model completed by training. The non-volatile computer-readable storage medium according to claim 16, when the computer-readable instructions are executed by the processor, the following steps are further executed: Dividing the set of entity elements by the extracted entity elements to obtain the accuracy rate of the entity element recognition model; Dividing the set of entity elements by the artificial entity elements to obtain the recall rate of the entity element recognition model; Calculate the product of the accuracy rate and the recall rate and the sum of the accuracy rate and the recall rate, divide the product by the sum value, and multiply the calculated result by the preset value as the first of the entity element recognition model F value; According to the accuracy rate, the recall rate and the first F value of the entity element recognition model, the accuracy of the entity element recognition model is calculated. The non-volatile computer-readable storage medium according to claim 15, when the computer-readable instructions are executed by the processor, the following steps are further executed: Extracting the above-mentioned training corpus through the label recognition model to obtain the corresponding extracted label; Determining that the tags in the manual tags that overlap with the extracted tags are used as a tag set; Calculating the accuracy of the label recognition model according to the label set, the artificial label, and the extracted label; The label recognition model whose accuracy exceeds the preset second accuracy is used as the trained label recognition model. The non-volatile computer-readable storage medium according to claim 18, when the computer-readable instructions are executed by the processor, the following steps are further executed: Dividing the label set by the extracted label to obtain the accuracy rate of the label recognition model; Dividing the label set by the artificial label to obtain the recall rate of the label recognition model; Calculate the product of the accuracy rate and the recall rate and the sum of the accuracy rate and the recall rate, divide the product by the sum value, and multiply the calculated result by the preset value as the second F of the label recognition model value; According to the accuracy rate, the recall rate and the second F value of the tag recognition model, the accuracy of the tag recognition model is calculated. The non-volatile computer-readable storage medium according to claim 15, when the computer-readable instructions are executed by the processor, the following steps are further executed: Using artificial entity elements and artificial tags as inputs of a preset TransE algorithm, so that the artificial entity elements and the artificial tags are embedded in a low-dimensional vector space to generate a corresponding vector template; The vector template is stored in the input graph database to construct a corresponding knowledge graph.