WO2021017721A1

WO2021017721A1 - Intelligent question answering method and apparatus, medium and electronic device

Info

Publication number: WO2021017721A1
Application number: PCT/CN2020/098948
Authority: WO
Inventors: 张国辉; 钱柏丞
Original assignee: 平安科技（深圳）有限公司
Priority date: 2019-08-01
Filing date: 2020-06-29
Publication date: 2021-02-04
Also published as: CN110647614A; CN110647614B

Abstract

An intelligent question answering method and apparatus, a medium and an electronic device. The method comprises: preprocessing a question inputted by a user and a plurality of question templates in a preset template library to obtain keywords of the question and question templates respectively; inputting question and template keyword group pairs composed of the keywords of the question and the keyword of each question template to an established semantic parser that comprises a multi-layer word similarity determination model to obtain the degree of similarity between the question and each question template outputted by the semantic parser; acquiring a standard question template from all question templates in the preset template library according to the degree of similarity; and determining an answer to a question corresponding to the standard question template and outputting the answer to the question. The described method may obtain the answer that best matches the question of a user with respect to the question, which improves the efficiency and accuracy of question and answer matching, thereby improving the matching effect.

Description

Intelligent question answering method, device, medium and electronic equipment

Cross references to related applications

This application claims the priority of a Chinese patent application filed with the Chinese Patent Office on August 1, 2019, the application number is 201910709165.9, and the invention title is "Intelligent Question Answering Methods, Devices, Media and Electronic Equipment", the entire contents of which are incorporated by reference In this application.

Technical field

This application relates to the field of artificial intelligence, and is specifically applied to the technical fields of natural language processing and semantic analysis, and in particular to an intelligent question answering method, device, medium and electronic equipment.

Background technique

Intelligent question answering systems and chat robots have always been research hotspots in the field of natural language processing. These research directions are mainly to find ways to make computers make appropriate answers to user questions.

In the implementation of the prior art, in order to match the user's question with the answer, a separate language similarity model is usually used to match the question and the answer template for similarity. However, the inventor realizes that the questions input by the user are often very different, and the semantics may be very complicated, which brings a high degree of difficulty to the matching of the question and the answer, and often leads to low matching efficiency and accuracy of the question and answer, and the matching effect difference. Therefore, the prior art urgently needs a technical solution that improves the efficiency and accuracy of question and answer matching, thereby improving the matching effect.

Summary of the invention

In the field of artificial intelligence and semantic analysis technology, in order to solve the above technical problems, the purpose of this application is to provide an intelligent question answering method, device, medium and electronic equipment.

According to the first aspect of the present application, there is provided an intelligent question answering method, which includes:

Preprocessing the question sentence input by the user and multiple question sentence templates in the preset template library to obtain the keywords of the question sentence and the question sentence template respectively;

Input the question sentence and template keyword group pair composed of the keywords of the question sentence and the keywords of each question template into the established semantic parser including the multi-layer word similarity judgment model to obtain the The similarity between the question output from the semantic parser and each question template;

Obtaining a standard question template from all question templates in a preset template library according to the similarity;

Determine the answer to the question corresponding to the standard question template and output the answer to the question.

According to the second aspect of the present application, there is provided an intelligent question answering device, the device comprising:

The keyword acquisition module is configured to preprocess the question sentence input by the user and a plurality of question sentence templates in a preset template library to obtain the keywords of the question sentence and the question sentence template respectively;

The similarity acquisition module is configured to input the question sentence and template keyword group pair composed of the keywords of the question sentence and the keywords of each of the question sentence templates to the established similarity determination including multi-layer words The semantic parser of the model obtains the similarity between the question sentence output by the semantic parser and each question template;

A template obtaining module configured to obtain a standard question template from all question templates in a preset template library according to the similarity;

The output module is configured to determine a question answer corresponding to the standard question template and output the question answer.

According to a third aspect of the present application, there is provided an electronic device, the electronic device including:

Processor; memory, the memory stores computer-readable instructions, when the computer-readable instructions are executed by the processor, the method as described above is implemented.

According to the fourth aspect of the present application, a computer-readable storage medium is provided, which stores computer program instructions, and when the computer program instructions are executed by a computer, the computer executes the aforementioned method

The technical solutions provided by the embodiments of the present application may include the following beneficial effects:

The intelligent question answering method provided by this application includes the following steps: preprocessing the question sentence input by the user and a plurality of question sentence templates in a preset template library to obtain the keywords of the question sentence and the question sentence template respectively ; Input the question sentence and template keyword group pair based on the keywords of the question sentence and the keywords of each of the question sentence templates into the established semantic parser including the multi-layer word similarity judgment model to obtain The degree of similarity between the question sentence output by the semantic parser and each of the question sentence templates; the standard question template is obtained from all the question sentence templates in the preset template library according to the similarity; State the answer to the question corresponding to the standard question template and output the answer to the question.

In this method, after obtaining the keywords of the question input by the user and the preset question template respectively, by constructing a keyword group pair and inputting it to the already constructed semantic parser including a multi-layer word similarity judgment model, A more accurate output result of the similarity between the question and the question model can be obtained, and then the question template that best matches the question entered by the user can be obtained according to the result, so that the best match to the question can be obtained for the user’s question The answer improves the efficiency and accuracy of question and answer matching, thereby improving the matching effect.

It should be understood that the above general description and the following detailed description are only exemplary and cannot limit the application.

Description of the drawings

The drawings here are incorporated into the specification and constitute a part of the specification, show embodiments that conform to the application, and are used together with the specification to explain the principle of the application.

Fig. 1 is a schematic diagram showing an application scenario of an intelligent question answering method according to an exemplary embodiment;

Fig. 2 is a flowchart showing an intelligent question answering method according to an exemplary embodiment;

FIG. 3 is a flowchart of the similarity between question sentences and question sentence templates output by the semantic parser according to an embodiment shown in the embodiment corresponding to FIG. 2;

FIG. 4 is a flowchart showing details of step 330 in an embodiment according to the embodiment corresponding to FIG. 3;

Fig. 5 is a schematic diagram showing a logical framework of a semantic parser outputting an inferred similarity matrix according to an exemplary embodiment;

Fig. 6 is a block diagram showing an intelligent question answering device according to an exemplary embodiment;

Fig. 7 is a block diagram showing an example of an electronic device implementing the above intelligent question answering method according to an exemplary embodiment;

Fig. 8 shows a computer-readable storage medium for implementing the above intelligent question answering method according to an exemplary embodiment.

Detailed ways

Here, exemplary embodiments will be described in detail, and examples thereof are shown in the accompanying drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings indicate the same or similar elements. The implementation manners described in the following exemplary embodiments do not represent all implementation manners consistent with the present application. On the contrary, they are only examples of devices and methods consistent with some aspects of the application as detailed in the appended claims.

In addition, the drawings are only schematic illustrations of the application, and are not necessarily drawn to scale. The same reference numerals in the figures denote the same or similar parts, and thus their repeated description will be omitted. Some of the block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities.

This application first provides an intelligent question answering method. Intelligent Q&A is the process of outputting corresponding answers based on questions. Regardless of whether it is a question or an answer, it can be a symbol or text in essence, and is often processed as a text in the computer field. In terms of specific expression, the expression of the question and the answer can be the same or different. Questions and answers can exist in various forms of expression such as phrases, phrases, sentences, paragraphs, and articles. Typically, questions can exist in the form of sentences, and answers can exist in the form of sentences or paragraphs. The content of the question is generally the doubt or question to be answered, and the answer is the answer or feedback to the corresponding question. Under normal circumstances, for most of the problems that people encounter, there are corresponding relatively correct answers that have been scientifically or practically tested. People explore the unknown and want to get answers to questions. Therefore, it is necessary to provide people with accurate answers based on people’s problems. The answer, since the question and the corresponding answer are completely two-dimensional content, how to accurately match the answer with the corresponding question is exactly the problem to be solved by the intelligent question and answer method provided by this application, through the intelligent question and answer method provided by this application , Can achieve an accurate match between the answer and the corresponding question.

The implementation terminal of this application can be any device with computing and processing functions. The device can be connected to an external device to receive or send data. Specifically, it can be a portable mobile device, such as a smart phone, a tablet computer, a notebook computer, or a PDA ( Personal Digital Assistant), etc., can also be fixed devices, such as computer equipment, field terminals, desktop computers, servers, workstations, etc., or a collection of multiple devices, such as the physical infrastructure of cloud computing.

Optionally, the implementation terminal of this application may be a server or a physical infrastructure of cloud computing.

Fig. 1 is a schematic diagram showing an application scenario of an intelligent question answering method according to an exemplary embodiment. As shown in Figure 1, it includes a server 110, a database 120, and a user terminal 130. The database 120 and the user terminal 130 are respectively connected to the server 110 through a communication link. In this embodiment, the server 110 is the implementation terminal of the application. The user terminal 130 can also be any device with computing and processing functions. It can be the same type of terminal as the implementation terminal of this application, or it can be a different type of terminal, and can be the same terminal as the implementation terminal of this application, or it can For different terminals. When the user needs an answer to a search question, he can first input the question he wants to find an answer through the input device (such as keyboard, mouse, touch screen, etc.) of the user terminal 130. The question may be sent to the server 110 through the user terminal 130, and The server 110 finds an answer that matches the question. The database 120 stores multiple question templates and question answers corresponding to the question templates. When the server 110 wants to find an answer that matches the target question, it can first find the question template that best matches the target question in the database 120 according to the target question. , And then use the answer to the question corresponding to the found question template as the answer that matches the target question. Finally, the server 110 may return the answer to the user terminal 130 to provide the user of the user terminal 130 with a corresponding answer.

It is worth mentioning that although in this embodiment, the implementation terminal of this application is a server, and the answer to the question corresponding to the question template is stored in the database, in other embodiments or specific applications, various terminals can be used as needed. It is selected as the implementation terminal of this application, and the question template and the answer to the question can be stored on any same or different terminal. This application does not make any limitation on this, and the protection scope of this application should not be restricted in any way.

Fig. 2 is a flow chart showing a method for intelligent question answering according to an exemplary embodiment. As shown in Figure 2, the following steps can be included:

Step 210: Preprocessing the question sentence input by the user and a plurality of question sentence templates in a preset template library to obtain keywords of the question sentence and the question sentence template respectively.

As mentioned above, the expression form of question sentence can be in various forms such as phrase, phrase, sentence, paragraph.

The question template can exist in the same form as the question sentence, or in a different form, and the question template can be matched with the question sentence.

For example, the question can be "Excuse me, what is the phone number of Zhang San", then the question template corresponding to the question can be: the mobile phone number of <name entity>.

In one embodiment, the preprocessing is to obtain the keywords of the question sentence or question template.

Keywords are words, words or phrases that record the core content of a question or question template.

In one embodiment, the preprocessing of the question sentence input by the user and the multiple question sentence templates in the preset template library to obtain the keywords of the question sentence and the question sentence template respectively includes:

Separate the question sentence input by the user and multiple question sentence templates in the preset template library;

For the question entered by the user and each question template in the preset template library, stop words are removed from the words cut from the question entered for the user and the question template in the preset template library , Get the key words of the question and the question template.

Word cutting of question sentence and question sentence template is the process of dividing question sentence and question sentence model into words.

The advantage of this embodiment is that the keywords are obtained by removing only the stop words in the divided words, which ensures that the obtained keywords can represent more question sentences or question template information.

In one embodiment, the question sentence input by the user and the multiple question sentence templates in the preset template library are segmented by calling a preset word segmentation interface.

Stop words mean that certain words or words will be automatically filtered before or after processing natural language data (or text) in order to save storage space and improve search efficiency in information retrieval.

For example, in English, words such as "the", "at", and "is" can be removed as stop words; in Chinese, words such as "的", "是", "在" can be used as stop words Remove.

For the question input by the user or each question template in the preset template library, the words that are cut from the question input for the user or the question template in the preset template library are obtained in advance. Set the words in the keyword library to obtain the keywords of the question sentence or the question sentence template.

The advantage of this embodiment is that the range of keywords of the question or question template obtained is limited by the preset keyword library, so that the keywords of the obtained question and the question template can be kept relatively high the quality of.

In one embodiment, for the question input by the user or each question template in the preset template library, the question template in the question input for the user or the preset template library is selected. From the obtained words, obtaining the words existing in the preset keyword library, and obtaining the keywords of the question sentence or the question sentence template may include:

For the question entered by the user or each question template in the preset template library, for each word in the question entered for the user or the question template in the preset template library , Judge whether the word exists in the preset keyword library; if so, use the word as the keyword of the question sentence or the question sentence template.

Step 220: Input the question sentence and template keyword group pair composed of the keywords of the question sentence and the keywords of each question template to the established semantic parser including the multi-layer word similarity judgment model To obtain the similarity between the question output by the semantic parser and each question template.

In an embodiment, the multi-layer word similarity judgment model in the semantic parser includes:

Jaccard index model layer, word2vec model layer, Glove model layer and C&W model layer.

The above-mentioned word similarity judgment model can output the similarity of question sentence and template keyword group pair, and one model layer is a word similarity judgment model.

Specifically, the Jaccard index model contained in the Jaccard index model layer can directly output the similarity, while the models in the word2vec model layer, the Glove model layer and the C&W model layer may not directly output the similarity, but output Word vectors. In addition to the model, these model layers can also include a similarity calculation layer, which can be used to calculate the similarity of word vectors, so that the model layer can output the similarity of the question sentence and the template keyword group pair. The similarity calculation layer The similarity of word vectors can be calculated based on cosine distance or Euclidean distance.

It should be noted that although in the above embodiment, the word similarity judgment model is a model layer, and some model layers may include the model and the similarity calculation layer, but here is only for ease of understanding and description, the model and similarity calculation The layers are collectively called the word similarity judgment model. In actual situations, the word similarity judgment model may not be a separate model, but a collection of units or components that can complete the function of word similarity judgment.

In one embodiment, in addition to the Jaccard index model layer, the word2vec model layer, the Glove model layer, and the C&W model layer, the semantic parser further includes: a word similarity judgment model layer in a specific field.

The word similarity judgment model in a specific field can be a model that can be calculated based on experience in certain fields. The specific implementation form can be arbitrary. For example, it can be specified by pre-establishing a word correlation database in a specific field. Judgment of similarity of words in the field.

The advantage of this embodiment is that by adding a layer of word similarity judgment model, the accuracy of the similarity between the question sentence output by the semantic parser and each question sentence template can be improved to a certain extent.

In an embodiment, the specific manner of the similarity between the question sentence output by the semantic parser and each question sentence template may be shown in FIG. 3. Fig. 3 is a flowchart of the semantic parser outputting question sentence and question sentence template similarity according to an embodiment of the embodiment corresponding to Fig. 2, as shown in Fig. 3, including the following steps:

Step 310: Obtain a keyword pair matrix corresponding to each of the question sentence and the template keyword group pair.

In an embodiment, each element in the keyword pair matrix is a combination of the keywords of the question sentence and the question template, and all elements in the keyword pair matrix contain the key of the question sentence and the question template. For all combinations of words, the number of keywords in the question sentence is the row width or column width of the keyword pair matrix. Correspondingly, the number of keywords in the question sentence template is the column width or row width of the keyword pair matrix. .

In an embodiment, the obtaining the keyword pair matrix corresponding to each of the question sentence and the template keyword group pair includes:

Starting from the first question keyword in the question keyword group in the question sentence and the template keyword group pair, according to the order of the question sentence keywords in the question sentence keyword group, aiming at the question keyword group For each question keyword, in the template keyword group of the question sentence and the template keyword group, start from the first template keyword in the order of the template keywords in the template keyword group, get the template keywords in turn, and The obtained template keyword and the question keyword form a keyword pair;

Acquiring the number of question keywords in the question keyword group in the question sentence and the template keyword group pair;

The keyword pairs are sorted according to the generation order, starting from the first keyword pair, each time the number of keyword pairs that have not been obtained are obtained in the stated order, as a keyword pair matrix In the nth row in, where n is the number of times of obtaining the number of keyword pairs.

Step 320: Use each layer of word similarity judgment model in the semantic parser to perform word similarity judgment on each keyword pair matrix, and obtain the word similarity judgment model for each layer of word similarity judgment for each question. The similarity matrix of the output keyword pairs of sentence and template keyword group.

As mentioned earlier, the word similarity judgment model can be the Jaccard index model layer, the word2vec model layer, the Glove model layer, the C&W model layer, the word similarity judgment model layer in a specific field, etc. The word similarity judgment model can be used for key The similarity is calculated for each keyword pair in the word pair matrix to obtain the similarity matrix.

Step 330: For each of the question sentence and the template keyword group pair, the question sentence and the template are obtained based on the similarity matrix of the keyword pair of the question sentence and the template keyword group pair output by the word similarity judgment model of each layer The inferred similarity matrix of the keyword pair.

The inferred similarity matrix is a combination of the similarity matrices output by the word similarity judgment model of each layer. By integrating the output results of the word similarity judgment model of each layer, the obtained inferred similarity matrix can be more reliable and accurate. .

In an embodiment, each element in the similarity matrix corresponds to a keyword pair in a question sentence and a template keyword group pair, and for each of the question sentence and template keyword group pair, based on The similarity matrix of the keyword pairs of the question sentence and the template keyword group pair output by the layer word similarity judgment model, to obtain the inferred similarity matrix of the question sentence and the template keyword group pair, including:

For each question sentence and template keyword group pair, for each keyword pair corresponding to the question sentence and template keyword group pair, the similarity matrix of the keyword pairs output by the word similarity judgment model of each layer is compared with This keyword sorts the corresponding elements from large to small;

For each question sentence and template keyword group pair, for each keyword pair corresponding to the question sentence and template keyword group pair, in the similarity matrix of the keyword pairs output by the word similarity judgment model of each layer and Among the elements corresponding to the keyword pairs, the elements corresponding to the first predetermined number of keyword pairs in the first order are obtained as the inferred similarity matrix obtaining elements;

For each pair of the question and template keyword, elements are obtained based on the obtained inferred similarity matrix to obtain an inferred similarity matrix.

The inferred similarity matrix acquisition element is used to obtain elements from which to construct the inferred similarity matrix.

In one embodiment, for each pair of the question and template keyword, obtaining elements based on the obtained inferred similarity matrix to obtain an inferred similarity matrix includes:

For each keyword pair corresponding to the question and template keyword group pair, determine the variance of each inferred similarity matrix to obtain elements;

In the case where the variance is greater than the predetermined variance threshold, obtain the maximum value of the element corresponding to the keyword pair;

In the case that the variance is not greater than the predetermined variance threshold, any one element is selected from the elements corresponding to the keyword pair in the second predetermined number in the first order, and the second predetermined number is less than the first predetermined number;

For the maximum value of the corresponding elements obtained for each keyword pair and/or arbitrary elements, the similarity matrix is formed according to the order of each corresponding element in the similarity matrix, as the inference similarity of the question and the template keyword group pair Degree matrix.

Since the value of each keyword to the corresponding element may not completely accurately reflect the value of the element in the final inferred similarity matrix, especially when the value of each element corresponding to the same keyword is not much different, use It may be unreasonable for keywords to construct an inferred similarity matrix based on the maximum value of corresponding elements. Therefore, the advantage of this embodiment is that when the variance is not greater than the predetermined variance threshold, that is, when the value of each element corresponding to the same keyword is sufficiently similar, the value of each element corresponding to the keyword is ranked sufficiently reliable. Any one of the previous elements is used to construct the inferred similarity matrix of the question and template keyword group pair, which improves the randomness and fairness of the constructed inferred similarity matrix, and may improve the acquisition to a certain extent The accuracy of the inferred similarity matrix.

For each keyword pair corresponding to the question and template keyword group pair, determine the average value of each inferred similarity matrix to obtain elements;

For each keyword pair corresponding to the question and template keyword group pair, obtain the inferred similarity matrix corresponding to the keyword pair to obtain the maximum value of the element and the ratio of the difference between the average value and the maximum value;

In the case that the ratio is less than the predetermined ratio threshold, obtain the maximum value of the corresponding element for the keyword;

In the case that the ratio is greater than or equal to a predetermined ratio threshold, obtaining an average value of elements for each inferred similarity matrix corresponding to the keyword pair;

Obtain the maximum value of the corresponding elements for each keyword pair and/or the average value of the inferred similarity matrix, and form the similarity matrix according to the order of each corresponding element in the similarity matrix, as the key to the question and template Inferred similarity matrix of phrase pairs.

The average value of the elements obtained by the inferred similarity matrix reflects the central tendency of the elements obtained by the inferred similarity matrix. In this embodiment, the difference between the average value of the elements obtained by the inferred similarity matrix and the maximum value of the corresponding element of the keyword pair is different. In large cases, the similarity matrix is constructed by obtaining the average value of each inferred similarity matrix, so that the elements in the similarity matrix obtained by obtaining the average value of the elements based on each inferred similarity matrix can better reflect the similarity of words in each layer The judgment result of the degree judgment model improves the fairness of constructing the inferred similarity matrix. At the same time, because the keyword pair corresponding to the inferred similarity matrix is selected, the maximum value of the element and the difference between the average value and the maximum value are obtained. The ratio is used as a condition for determining whether to construct the similarity matrix by obtaining the average value of the elements obtained by each inferred similarity matrix. Whether the maximum value of the elements obtained by the inferred similarity matrix is an extreme value, this condition can be passed or not to a certain extent. Obtain each inferred similarity matrix to obtain the average value of the elements to construct the similarity matrix for good distinction, which improves the generalization ability and universality of the condition.

Fig. 5 is a schematic diagram showing a logical framework of a semantic parser outputting an inferred similarity matrix according to an exemplary embodiment. As shown in FIG. 5, it includes a keyword pair matrix 510, a first-level word similarity determination model 520, a second-level word similarity determination model 530, a third-level word similarity determination model 540, a similarity matrix 550, and inferred similarity The degree matrix 560, in which the three-layer word similarity judgment model belongs to the semantic parser 500, the similarity matrix pointed by the arrow pointed by the word similarity judgment model of each layer is the similarity output of the word similarity judgment model of each layer The inferred similarity matrix 560 is obtained based on the similarity matrix output by the word similarity judgment model of each layer.

Step 340, for each question sentence and template keyword group pair, obtain the question sentence and question sentence corresponding to the question sentence and the template keyword group pair according to each similarity in the inferred similarity matrix of the question sentence and the template keyword group pair The similarity of the template.

In one embodiment, for each question sentence and template keyword group pair, according to each similarity in the inferred similarity matrix of the question sentence and the template keyword group pair, obtain the corresponding question sentence and the template keyword group pair The similarity between question and question template, including:

According to the similarity in the inferred similarity matrix of the question sentence and the template keyword group pair, the similarity of the question sentence and the question template corresponding to the question sentence and the template keyword group pair is obtained by the following formula:

Wherein, S1 and S2 are the question sentence and the question sentence template respectively, Sim(S1, S2) is the similarity between the question sentence and the question sentence template, and w _i is the key word in the question sentence , And w _j is the keyword in the question template, maxSim(w _j ) is the maximum value of the similarity between each keyword in the question template and the keyword w _i in the question, maxSim (w _i ) is the maximum value of the similarity between each keyword in the question sentence and the keyword w _j in the question sentence, idf(w _i ) is the reverse of the keyword w _{i in} the question sentence File frequency, idf(w _j ) is the reverse file frequency of the keyword w _{j in} the question sentence, D is the number of all question sentence templates,

Is the number of question template for the keyword w _i in the question sentence.

Next, referring to FIG. 2 continuously, the following steps are included after step 220:

Step 230: Obtain a standard question template from all question templates in a preset template library according to the similarity.

In one embodiment, the obtaining the standard question template from all question templates in the preset template library according to the similarity includes: obtaining the standard question template from all question templates in the preset template library The question template with the greatest similarity is used as the standard question template.

In one embodiment, obtaining a standard question template from all question templates in a preset template library according to the similarity includes: obtaining all question templates in the preset template library according to the similarity from Sorting from large to small; obtaining the predetermined number of question template in the order of similarity from all question templates in the preset template library, as the standard question template.

In one embodiment, the obtaining standard question template from all question template in a preset template library according to the similarity includes:

Sort all question templates in the preset template library from large to small according to the similarity;

In the case that the number of question templates whose similarity is greater than the preset similarity threshold is greater than or equal to the predetermined number threshold, the question templates ranked less than or equal to the predetermined number threshold are used as standard question templates;

In the case that the number of question template with the similarity greater than the preset similarity threshold is less than the predetermined number threshold, the question template with the similarity greater than the preset similarity threshold is used as the standard question template.

The advantage of this embodiment is that by limiting the number of obtained standard templates, the efficiency of outputting question answers can be improved. At the same time, by using the preset similarity threshold as the basis for selecting standard question templates, the standard question templates obtained are improved. Accuracy.

Step 240: Determine the question answer corresponding to the standard question template and output the question answer.

In one embodiment, the question answer corresponding to each question template is stored in the preset database. The determining the question answer corresponding to the standard question template and outputting the question answer includes: querying The preset database obtains the answer to the question corresponding to the standard question template and outputs the answer to the question.

In one embodiment, the determining the question answer corresponding to the standard question template and outputting the question answer includes: obtaining the answer corresponding to the standard question template by calling a preset question answer query interface The answer to the question and the answer to the question are output.

In one embodiment, the question answer data is stored in the form of RDF (Resource Description Framework) data, and the determining the question answer corresponding to the standard question template and outputting the question answer includes:

The question answer data stored in the form of RDF data is queried through the SPARQL sentence to obtain the question answer corresponding to the standard question template and output the question answer.

In one embodiment, determining the question answer corresponding to the standard question template and outputting the question answer includes:

Determine the question answer corresponding to the standard question template; obtain the question answer template; inject the question answer into the question answer template for output.

To sum up, according to the intelligent question and answer method provided by the embodiment of FIG. 2, after obtaining the user input question and the keywords of the preset question template, the keyword group pair is constructed and input to the already constructed including The semantic parser of the multi-layer word similarity judgment model can obtain a more accurate output result of the similarity between the question and the question model, and then according to the result, the question template that best matches the question input by the user can be obtained, thereby The answer that best matches the question can be obtained for the user’s question, which improves the efficiency and accuracy of question and answer matching, thereby improving the matching effect.

FIG. 4 is a flowchart showing details of step 330 in an embodiment according to the embodiment corresponding to FIG. 3. As shown in Figure 4, the following steps can be included:

Step 331: For each question sentence and template keyword group pair, for each keyword pair corresponding to the question sentence and template keyword group pair, obtain the similarity of the keyword pairs output by the word similarity judgment model of each layer The maximum value of the element corresponding to the keyword pair in the matrix.

In one embodiment, the maximum value is obtained through the first round of bubble sorting.

In step 332, the maximum value of the corresponding element obtained for each keyword pair corresponding to the question sentence and the template keyword group pair is formed into a similarity matrix according to the order of each corresponding element in the similarity matrix, as the Inferred similarity matrix of question sentence and template keyword group pair.

Continuing to refer to FIG. 5, each element in the inferred similarity matrix 560 is the maximum value of the element corresponding to the keyword pair in the similarity matrix of the keyword pair output by the word similarity judgment model of each layer, for example, for For the keyword pair A1B1, the similarity matrix output by the three-layer word similarity judgment model and the corresponding elements of the keyword pair are 0.85, 0.84, and 0.88, respectively, and the element corresponding to the keyword pair in the inferred similarity matrix is The maximum value of the three is 0.88.

The advantage of this embodiment is that the acquisition of elements in the inferred similarity matrix has fair standards, and the accuracy of the inferred similarity matrix is guaranteed.

This application also provides an intelligent question answering device. The following are device embodiments of this application.

Fig. 6 is a block diagram showing an intelligent question answering device according to an exemplary embodiment. As shown in FIG. 6, the apparatus 600 includes:

The keyword acquisition module 610 is configured to preprocess the question input by the user and a plurality of question template in a preset template library to obtain the keywords of the question sentence and the question template respectively;

The similarity acquisition module 620 is configured to input the question sentence and the template keyword group pair composed of the keywords of the question sentence and the keywords of each of the question sentence templates into the established similarity degree including multiple layers of words Determine the semantic parser of the model to obtain the similarity between the question sentence output by the semantic parser and each question template;

The template obtaining module 630 is configured to obtain a standard question template from all question templates in a preset template library according to the similarity;

The output module 640 is configured to determine the answer to the question corresponding to the standard question template and output the answer to the question.

According to the third aspect of the present application, there is also provided an electronic device capable of implementing the above method.

Those skilled in the art can understand that various aspects of the present application can be implemented as a system, method, or program product. Therefore, each aspect of the present application can be specifically implemented in the following forms, namely: complete hardware implementation, complete software implementation (including firmware, microcode, etc.), or a combination of hardware and software implementations, which can be collectively referred to herein as "Circuit", "Module" or "System".

The electronic device 700 according to this embodiment of the present application will be described below with reference to FIG. 7. The electronic device 700 shown in FIG. 7 is only an example, and should not bring any limitation to the function and use scope of the embodiments of the present application.

As shown in FIG. 7, the electronic device 700 is represented in the form of a general computing device. The components of the electronic device 700 may include, but are not limited to: the aforementioned at least one processing unit 710, the aforementioned at least one storage unit 720, and a bus 730 connecting different system components (including the storage unit 720 and the processing unit 710).

Wherein, the storage unit stores program code, and the program code can be executed by the processing unit 710, so that the processing unit 710 executes the various exemplary methods described in the "Embodiment Method" section of this specification. Implementation steps.

The storage unit 720 may include a readable medium in the form of a volatile storage unit, such as a random access storage unit (RAM) 721 and/or a cache storage unit 722, and may also optionally include a read-only storage unit (ROM) 723.

The storage unit 720 may also include a program/utility tool 724 having a set of (at least one) program module 725. Such program module 725 includes but is not limited to: an operating system, one or more application programs, other program modules, and program data, Each of these examples or some combination may include the implementation of a network environment.

The bus 730 may represent one or more of several types of bus structures, including a storage unit bus or a storage unit controller, a peripheral bus, a graphics acceleration port, a processing unit, or a local area using any bus structure among multiple bus structures. bus.

The electronic device 700 can also communicate with one or more external devices 900 (such as keyboards, pointing devices, Bluetooth devices, etc.), and can also communicate with one or more devices that enable users to interact with the electronic device 700, and/or communicate with Any device (such as a router, modem, etc.) that enables the electronic device 700 to communicate with one or more other computing devices. This communication can be performed through an input/output (I/O) interface 750. In addition, the electronic device 700 may also communicate with one or more networks (for example, a local area network (LAN), a wide area network (WAN), and/or a public network, such as the Internet) through the network adapter 760. As shown in the figure, the network adapter 760 communicates with other modules of the electronic device 700 through the bus 730. It should be understood that although not shown in the figure, other hardware and/or software modules can be used in conjunction with the electronic device 700, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives And data backup storage system, etc.

Through the description of the foregoing embodiments, those skilled in the art can easily understand that the exemplary embodiments described herein can be implemented by software, or can be implemented by combining software with necessary hardware. Therefore, the technical solution according to the embodiments of the present application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (can be a CD-ROM, U disk, mobile hard disk, etc.) or on the network , Including several instructions to make a computing device (which may be a personal computer, server, terminal device, or network device, etc.) execute the method according to the embodiment of the present application.

According to the fourth aspect of the present application, there is also provided a computer-readable storage medium on which is stored a program product capable of implementing the above method of this specification. In some possible implementation manners, various aspects of the present application can also be implemented in the form of a program product, which includes program code. When the program product runs on a terminal device, the program code is used to enable the The terminal device executes the steps according to various exemplary embodiments of the present application described in the above-mentioned "Exemplary Method" section of this specification.

Referring to FIG. 8, a program product 800 for implementing the above method according to an embodiment of the present application is described. It can adopt a portable compact disk read-only memory (CD-ROM) and include program code, and can be stored in a terminal device, For example, running on a personal computer. However, the program product of this application is not limited to this. In this document, the readable storage medium can be any tangible medium that contains or stores a program, and the program can be used by or combined with an instruction execution system, device, or device.

The program product can use any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. The readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or device, or a combination of any of the above. More specific examples (non-exhaustive list) of readable storage media include: electrical connections with one or more wires, portable disks, hard disks, random access memory (RAM), read only memory (ROM), erasable Type programmable read only memory (EPROM or flash memory), optical fiber, portable compact disk read only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination of the above.

The computer-readable signal medium may include a data signal propagated in baseband or as a part of a carrier wave, and readable program code is carried therein. This propagated data signal can take many forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the foregoing. The readable signal medium may also be any readable medium other than a readable storage medium, and the readable medium may send, propagate, or transmit a program for use by or in combination with the instruction execution system, apparatus, or device. The computer-readable storage medium may be non-volatile or volatile.

The program code contained on the readable medium can be transmitted by any suitable medium, including but not limited to wireless, wired, optical cable, RF, etc., or any suitable combination of the foregoing.

The program code used to perform the operations of this application can be written in any combination of one or more programming languages. The programming languages include object-oriented programming languages—such as Java, C++, etc., as well as conventional procedural Programming language-such as "C" language or similar programming language. The program code can be executed entirely on the user's computing device, partly on the user's device, executed as an independent software package, partly on the user's computing device and partly executed on the remote computing device, or entirely on the remote computing device or server Executed on. In the case of a remote computing device, the remote computing device can be connected to a user computing device through any kind of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computing device (for example, using Internet service providers) Business to connect via the Internet).

In addition, the above-mentioned drawings are only schematic illustrations of the processing included in the method according to the exemplary embodiments of the present application, and are not intended for limitation. It is easy to understand that the processing shown in the above drawings does not indicate or limit the time sequence of these processings. In addition, it is easy to understand that these processes can be executed synchronously or asynchronously in multiple modules, for example.

It should be understood that the present application is not limited to the precise structure that has been described above and shown in the drawings, and various modifications and changes can be performed without departing from its scope. The scope of the application is only limited by the appended claims.

Claims

An intelligent question answering method, wherein the method includes:

Preprocessing the question sentence input by the user and multiple question sentence templates in the preset template library to obtain the keywords of the question sentence and the question sentence template respectively;

Input the question sentence and template keyword group pair composed of the keywords of the question sentence and the keywords of each question template into the established semantic parser including the multi-layer word similarity judgment model to obtain the The similarity between the question output from the semantic parser and each question template;

Obtaining a standard question template from all question templates in the preset template library according to the similarity;

Determine the answer to the question corresponding to the standard question template and output the answer to the question.
4. The method according to claim 1, wherein the semantic parser is input to the semantic parser according to the question composed of the keywords of the question sentence and the keywords of each question template To output the similarity between the question sentence and each question sentence template with the template keyword group:

Acquire the keyword pair matrix corresponding to each of the question sentence and the template keyword pair;

Each layer of word similarity judgment model in the semantic parser is used to make word similarity judgment for each keyword pair matrix, and each layer of word similarity judgment model is obtained for each question and template. The similarity matrix of the keyword pairs output by the keyword group;

For each of the question sentence and template keyword group pair, the question sentence and the template keyword group pair are obtained based on the similarity matrix between the question sentence and the template keyword group pair output by the word similarity judgment model of each layer Inferred similarity matrix;

For each question sentence and template keyword group pair, according to each similarity in the inferred similarity matrix of the question sentence and the template keyword group pair, obtain the similarity of the question sentence and the question sentence template corresponding to the question sentence and the template keyword group pair degree.
3. The method according to claim 2, wherein said obtaining a keyword pair matrix corresponding to each of said question sentence and template keyword group pair comprises:

Starting from the first question keyword in the question keyword group in the question sentence and the template keyword group pair, according to the order of the question sentence keywords in the question sentence keyword group, aiming at the question keyword group For each question keyword, in the template keyword group of the question sentence and the template keyword group, start from the first template keyword in the order of the template keywords in the template keyword group, get the template keywords in turn, and The obtained template keyword and the question keyword form a keyword pair;

Acquiring the number of question keywords in the question keyword group in the question sentence and the template keyword group pair;

The keyword pairs are sorted according to the generation order, starting from the first keyword pair, each time the number of keyword pairs that have not been obtained are obtained in the stated order, as a keyword pair matrix In the nth row in, where n is the number of times of obtaining the number of keyword pairs.
The method according to claim 2, wherein each element in the similarity matrix corresponds to a keyword pair in a question sentence and a template keyword group pair, and each of the question sentence and template keyword group Yes, based on the similarity matrix of the keyword pair between the question and the template keyword set output by the word similarity judgment model of each layer, the inferred similarity matrix of the question and the template keyword set is obtained, including:

For each question sentence and template keyword group pair, for each keyword pair corresponding to the question sentence and template keyword group pair, obtain the similarity matrix of the keyword pairs output by the word similarity judgment model of each layer. The maximum value of the corresponding element of the keyword;

The maximum value of the corresponding element obtained for each keyword pair corresponding to the question and the template keyword group pair is formed into a similarity matrix according to the ordering of each corresponding element in the similarity matrix, as the question sum Inferred similarity matrix of template keyword group pairs.
The method according to claim 2, wherein each element in the similarity matrix corresponds to a keyword pair in a question sentence and a template keyword group pair, and each of the question sentence and template keyword group Yes, based on the similarity matrix of the keyword pair between the question and the template keyword set output by the word similarity judgment model of each layer, the inferred similarity matrix of the question and the template keyword set is obtained, including:

For each question sentence and template keyword group pair, for each keyword pair corresponding to the question sentence and template keyword group pair, the similarity matrix of the keyword pairs output by the word similarity judgment model of each layer is compared with This keyword sorts the corresponding elements from large to small;

For each question sentence and template keyword group pair, for each keyword pair corresponding to the question sentence and template keyword group pair, in the similarity matrix of the keyword pairs output by the word similarity judgment model of each layer and Among the elements corresponding to the keyword pairs, the elements corresponding to the first predetermined number of keyword pairs in the first order are obtained as the inferred similarity matrix obtaining elements;

For each pair of the question and template keyword, elements are obtained based on the obtained inferred similarity matrix to obtain an inferred similarity matrix.
The method according to claim 2, wherein, for each question sentence and template keyword group pair, the question sentence and the template are obtained according to each similarity in the inferred similarity matrix of the question sentence and the template keyword group pair The similarity of the keyword group to the corresponding question sentence and question template, including:

According to the similarity in the inferred similarity matrix of the question sentence and the template keyword group pair, the similarity of the question sentence and the question template corresponding to the question sentence and the template keyword group pair is obtained by the following formula:

Wherein, S1 and S2 are the question sentence and the question sentence template respectively, Sim(S1, S2) is the similarity between the question sentence and the question sentence template, and w i is the key word in the question sentence , And w j is the keyword in the question template, maxSim(w j ) is the maximum value of the similarity between each keyword in the question template and the keyword w i in the question, maxSim (w i ) is the maximum value of the similarity between each keyword in the question sentence and the keyword w j in the question sentence, idf(w i ) is the reverse of the keyword w i in the question sentence File frequency, idf(w j ) is the reverse file frequency of the keyword w j in the question sentence, D is the number of all question sentence templates,
Is the number of question template for the keyword w i in the question sentence.
The method according to claim 1, wherein said obtaining a standard question template from all question templates in said preset template library according to said similarity comprises:

Sorting all question templates in the preset template library in descending order of the similarity;

In a case where the number of question template whose similarity is greater than the preset similarity threshold is greater than or equal to a predetermined number threshold, the question template whose ranking is less than or equal to the predetermined number threshold is used as the standard question template;

In the case that the number of question templates whose similarity is greater than the preset similarity threshold is less than the predetermined number threshold, the question templates with the similarity greater than the preset similarity threshold are used as the Standard question template.
An intelligent question answering device, wherein the device includes:

The keyword acquisition module is configured to preprocess the question sentence input by the user and a plurality of question sentence templates in a preset template library to obtain the keywords of the question sentence and the question sentence template respectively;

The similarity acquisition module is configured to input the question sentence and template keyword group pair composed of the keywords of the question sentence and the keywords of each of the question sentence templates to the established similarity determination including multi-layer words The semantic parser of the model obtains the similarity between the question sentence output by the semantic parser and each question template;

A template obtaining module configured to obtain a standard question template from all question templates in the preset template library according to the similarity;

The output module is configured to determine a question answer corresponding to the standard question template and output the question answer.
An electronic device, wherein the electronic device includes:

processor;

A memory, the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, an intelligent question-and-answer method is implemented. The intelligent question-and-answer method specifically includes the following steps:

Preprocessing the question sentence input by the user and multiple question sentence templates in the preset template library to obtain the keywords of the question sentence and the question sentence template respectively;

Input the question sentence and template keyword group pair composed of the keywords of the question sentence and the keywords of each question template into the established semantic parser including the multi-layer word similarity judgment model to obtain the The similarity between the question output from the semantic parser and each question template;

Obtaining a standard question template from all question templates in the preset template library according to the similarity;

Determine the answer to the question corresponding to the standard question template and output the answer to the question.
9. The electronic device according to claim 9, wherein the semantic parser is inputted to the semantic parser according to the question consisting of the keywords of the question sentence and the keywords of each question template: Sentence and template keyword group pairs output the similarity between the question sentence and each question template:

Acquire the keyword pair matrix corresponding to each of the question sentence and the template keyword pair;

Each layer of word similarity judgment model in the semantic parser is used to make word similarity judgment for each keyword pair matrix, and each layer of word similarity judgment model is obtained for each question and template. The similarity matrix of the keyword pairs output by the keyword group;

For each of the question sentence and template keyword group pair, the question sentence and the template keyword group pair are obtained based on the similarity matrix between the question sentence and the template keyword group pair output by the word similarity judgment model of each layer Inferred similarity matrix;

For each question sentence and template keyword group pair, according to each similarity in the inferred similarity matrix of the question sentence and the template keyword group pair, obtain the similarity of the question sentence and the question sentence template corresponding to the question sentence and the template keyword group pair degree.
11. The electronic device according to claim 10, wherein said obtaining the keyword pair matrix corresponding to each of the question sentence and the template keyword group pair comprises:

Starting from the first question keyword in the question keyword group in the question sentence and the template keyword group pair, according to the order of the question sentence keywords in the question sentence keyword group, aiming at the question keyword group For each question keyword, in the template keyword group of the question sentence and the template keyword group, start from the first template keyword in the order of the template keywords in the template keyword group, get the template keywords in turn, and The obtained template keyword and the question keyword form a keyword pair;

Acquiring the number of question keywords in the question keyword group in the question sentence and the template keyword group pair;

The keyword pairs are sorted according to the generation order, starting from the first keyword pair, each time the number of keyword pairs that have not been obtained are obtained in the stated order, as a keyword pair matrix In the nth row in, where n is the number of times of obtaining the number of keyword pairs.
The electronic device according to claim 10, wherein each element in the similarity matrix corresponds to a keyword pair in a question sentence and a template keyword group pair, and the key to each question sentence and the template Phrase pair, based on the similarity matrix of the keyword pair of the question sentence and the template keyword group output from the word similarity judgment model of each layer, obtain the inferred similarity matrix of the question sentence and the template keyword group pair, including:

For each question sentence and template keyword group pair, for each keyword pair corresponding to the question sentence and template keyword group pair, obtain the similarity matrix of the keyword pairs output by the word similarity judgment model of each layer. The maximum value of the corresponding element of the keyword;

The maximum value of the corresponding element obtained for each keyword pair corresponding to the question and the template keyword group pair is formed into a similarity matrix according to the ordering of each corresponding element in the similarity matrix, as the question sum Inferred similarity matrix of template keyword group pairs.
The electronic device according to claim 10, wherein each element in the similarity matrix corresponds to a keyword pair in a question sentence and a template keyword group pair, and the key to each question sentence and the template Phrase pair, based on the similarity matrix of the keyword pair of the question sentence and the template keyword group output from the word similarity judgment model of each layer, obtain the inferred similarity matrix of the question sentence and the template keyword group pair, including:

For each question sentence and template keyword group pair, for each keyword pair corresponding to the question sentence and template keyword group pair, the similarity matrix of the keyword pairs output by the word similarity judgment model of each layer is compared with This keyword sorts the corresponding elements from large to small;

For each question sentence and template keyword group pair, for each keyword pair corresponding to the question sentence and template keyword group pair, in the similarity matrix of the keyword pairs output by the word similarity judgment model of each layer and Among the elements corresponding to the keyword pairs, the elements corresponding to the first predetermined number of keyword pairs in the first order are obtained as the inferred similarity matrix obtaining elements;

For each pair of the question and template keyword, elements are obtained based on the obtained inferred similarity matrix to obtain an inferred similarity matrix.
The electronic device according to claim 10, wherein, for each question sentence and template keyword group pair, the question sentence and the template keyword group are obtained according to each similarity in the inferred similarity matrix of the question sentence and the template keyword group pair. The similarity of the template keyword group to the corresponding question sentence and question sentence template, including:

Use the following formula to obtain the similarity between the question and the template keyword group corresponding to the question and the question template according to the similarity in the inferred similarity matrix of the question and the template keyword group pair:

Wherein, S1 and S2 are the question sentence and the question sentence template respectively, Sim(S1, S2) is the similarity between the question sentence and the question sentence template, and w i is the key word in the question sentence , And w j is the keyword in the question template, maxSim(w j ) is the maximum value of the similarity between each keyword in the question template and the keyword w i in the question, maxSim (w i ) is the maximum value of the similarity between each keyword in the question sentence and the keyword w j in the question sentence, idf(w i ) is the reverse of the keyword w i in the question sentence File frequency, idf(w j ) is the reverse file frequency of the keyword w j in the question sentence, D is the number of all question sentence templates,
Is the number of question template for the keyword w i in the question sentence.
9. The electronic device according to claim 9, wherein said obtaining a standard question template from all question templates in said preset template library according to said similarity comprises:

Sorting all question templates in the preset template library in descending order of the similarity;

In a case where the number of question template whose similarity is greater than the preset similarity threshold is greater than or equal to a predetermined number threshold, the question template whose ranking is less than or equal to the predetermined number threshold is used as the standard question template;

In the case that the number of question templates whose similarity is greater than the preset similarity threshold is less than the predetermined number threshold, the question templates with the similarity greater than the preset similarity threshold are used as the Standard question template.
A computer-readable storage medium, wherein computer program instructions are stored, and when the computer program instructions are executed by a computer, the computer executes an intelligent question and answer method. The intelligent question answer method specifically includes the following steps:

Preprocessing the question sentence input by the user and multiple question sentence templates in the preset template library to obtain the keywords of the question sentence and the question sentence template respectively;

Input the question sentence and template keyword group pair composed of the keywords of the question sentence and the keywords of each question template into the established semantic parser including the multi-layer word similarity judgment model to obtain the The similarity between the question output from the semantic parser and each question template;

Obtaining a standard question template from all question templates in the preset template library according to the similarity;

Determine the answer to the question corresponding to the standard question template and output the answer to the question.
The computer-readable storage medium according to claim 16, wherein the semantic parser is inputted to the semantic parser according to the keywords of the question sentence and the keywords of each question template The composed question sentence and template keyword group pair output the similarity between the question sentence and each question sentence template:

Acquire the keyword pair matrix corresponding to each of the question sentence and the template keyword pair;

Each layer of word similarity judgment model in the semantic parser is used to make word similarity judgment for each keyword pair matrix, and each layer of word similarity judgment model is obtained for each question and template. The similarity matrix of the keyword pairs output by the keyword group;

For each of the question sentence and template keyword group pair, the question sentence and the template keyword group pair are obtained based on the similarity matrix between the question sentence and the template keyword group pair output by the word similarity judgment model of each layer Inferred similarity matrix;

For each question sentence and template keyword group pair, according to each similarity in the inferred similarity matrix of the question sentence and the template keyword group pair, obtain the similarity of the question sentence and the question sentence template corresponding to the question sentence and the template keyword group pair degree.
18. The computer-readable storage medium according to claim 17, wherein said obtaining the keyword pair matrix corresponding to each of the question sentence and the template keyword group pair comprises:

Starting from the first question keyword in the question keyword group in the question sentence and the template keyword group pair, according to the order of the question sentence keywords in the question sentence keyword group, aiming at the question keyword group For each question keyword, in the template keyword group of the question sentence and the template keyword group, start from the first template keyword in the order of the template keywords in the template keyword group, get the template keywords in turn, and The obtained template keyword and the question keyword form a keyword pair;

Acquiring the number of question keywords in the question keyword group in the question sentence and the template keyword group pair;

The keyword pairs are sorted according to the generation order, starting from the first keyword pair, each time the number of keyword pairs that have not been obtained are obtained in the stated order, as a keyword pair matrix In the nth row in, where n is the number of times of obtaining the number of keyword pairs.
The computer-readable storage medium according to claim 17, wherein each element in the similarity matrix corresponds to a keyword pair in a question sentence and a template keyword group pair, and the And the template keyword group pair, based on the similarity matrix of the keyword pair between the question sentence and the template keyword group pair output by the word similarity judgment model of each layer, obtain the inferred similarity matrix of the question sentence and the template keyword group pair, including :

For each question sentence and template keyword group pair, for each keyword pair corresponding to the question sentence and template keyword group pair, obtain the similarity matrix of the keyword pairs output by the word similarity judgment model of each layer. The maximum value of the corresponding element of the keyword;

The maximum value of the corresponding element obtained for each keyword pair corresponding to the question and the template keyword group pair is formed into a similarity matrix according to the ordering of each corresponding element in the similarity matrix, as the question sum Inferred similarity matrix of template keyword group pairs.
The computer-readable storage medium according to claim 17, wherein each element in the similarity matrix corresponds to a keyword pair in a question sentence and a template keyword group pair, and the And the template keyword group pair, based on the similarity matrix of the keyword pair between the question sentence and the template keyword group pair output by the word similarity judgment model of each layer, obtain the inferred similarity matrix of the question sentence and the template keyword group pair, including :

For each question sentence and template keyword group pair, for each keyword pair corresponding to the question sentence and template keyword group pair, the similarity matrix of the keyword pairs output by the word similarity judgment model of each layer is compared with This keyword sorts the corresponding elements from large to small;

For each question sentence and template keyword group pair, for each keyword pair corresponding to the question sentence and template keyword group pair, in the similarity matrix of the keyword pairs output by the word similarity judgment model of each layer and Among the elements corresponding to the keyword pairs, the elements corresponding to the first predetermined number of keyword pairs in the first order are obtained as the inferred similarity matrix obtaining elements;

For each pair of the question and template keyword, elements are obtained based on the obtained inferred similarity matrix to obtain an inferred similarity matrix.