Question and answer corpus construction method and device, question and answer method, equipment and medium

Abstract

The embodiment of the application provides a kind of question and answer corpus construction method and device, question and answer method, equipment and medium, belong to the field of financial science and technology.The method comprises: obtaining historical dialogue data;Wherein, historical dialogue data includes historical question and answer data and dialogue identification data;According to dialogue identification data, data segmentation processing is carried out to historical question and answer data, and initial question and answer data is obtained;According to the first preset data volume, data volume screening is carried out to initial question and answer data, and candidate question and answer data is obtained;According to the preset target question model, content recognition is carried out to original question data, and the target question label of original question data is obtained;Wherein, target question label is used to indicate that the content category of original question data is question class or non-question class;According to target question label, candidate question and answer data is screened, and target question and answer data is obtained;According to target question and answer data, question and answer corpus is constructed.The embodiment of the application can improve the construction efficiency of question and answer corpus.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of financial technology, and in particular to a question and answer corpus construction method and device, a question and answer method, equipment and a medium. BACKGROUND

[0002] At present, in the field of financial technology, a question and answer corpus can be used to intelligently reply to user questions, such as intelligently replying to user questions about insurance consultation. In related technologies, a question and answer corpus is constructed or expanded by manual annotation. However, the above method requires a long manual annotation time, thereby affecting the construction efficiency of the question and answer corpus. Therefore, how to improve the construction efficiency of the question and answer corpus has become a technical problem to be solved. SUMMARY

[0003] The main purpose of the embodiments of the present application is to provide a question and answer corpus construction method and device, equipment and a medium, which aims to improve the construction efficiency of the question and answer corpus.

[0004] To achieve the above purpose, the first aspect of the embodiments of the present application provides a question and answer corpus construction method, which comprises:

[0005] Obtain historical dialogue data; wherein the historical dialogue data comprises historical question and answer data and dialogue identification data; wherein the dialogue identification data is used to identify the question and answer information of the historical question and answer data;

[0006] According to the dialogue identification data, the historical question and answer data is subjected to data segmentation processing to obtain initial question and answer data;

[0007] According to the first preset data amount, the initial question and answer data is subjected to data amount screening to obtain candidate question and answer data; wherein the candidate question and answer data comprises original question data;

[0008] According to a preset target question model, the original question data is subjected to content recognition to obtain a target question label of the original question data; wherein the target question label is used to indicate that the content category of the original question data is a question category or a non-question category;

[0009] According to the target question label, the candidate question and answer data is subjected to screening to obtain target question and answer data;

[0010] According to the target question and answer data, a question and answer corpus is constructed.

[0011] In some embodiments, the target question label comprises a question positive label, which is used to indicate that the content category of the original question data is a question category;

[0012] The target question data is obtained by screening the candidate question and answer data according to the target question label.

[0013] The target question data is obtained by screening the original question data according to the question label.

[0014] The target data amount is obtained by obtaining the data amount of the target question data in the candidate question and answer data.

[0015] If the target data amount is less than the second preset data amount, the candidate question and answer data is taken as the target question and answer data.

[0016] In some embodiments, the candidate question and answer data is obtained by performing data amount screening on the initial question and answer data according to a first preset data amount, and the method further comprises:

[0017] The initial semantic vector is obtained by performing semantic recognition on the initial question and answer data.

[0018] The first matching result is obtained by performing semantic matching on the initial semantic vector and a preset semantic vector.

[0019] The initial semantic vector matched by the first matching result is taken as a target semantic vector, and the key question and answer data is obtained by screening the initial question data according to the target semantic vector.

[0020] The candidate question and answer data is obtained by performing data amount screening on the key question and answer data according to the first preset data amount.

[0021] In some embodiments, the key question and answer data comprises key question data.

[0022] The candidate question and answer data is obtained by performing data amount screening on the key question and answer data according to the first preset data amount, and the method further comprises:

[0023] The question data amount is obtained by obtaining the question data of the key question data.

[0024] If the question data amount is less than the first preset data amount, the key question and answer data is taken as the candidate question and answer data.

[0025] In some embodiments, before the target question label of the original question data is obtained by performing content recognition on the original question data according to a preset target question model, the method further comprises training the target question model, and specifically comprises:

[0026] The sample question data and a sample question label of the sample question data are obtained, wherein the sample question label is used to represent that the content category of the sample question data is a question category or a non-question category.

[0027] According to the preset original question model, content recognition is performed on the sample question data to obtain an original question label; wherein the original question label is used to represent that the data category of the sample question data is a question category or a non-question category;

[0028] According to the sample question label and the original question label, parameter adjustment is performed on the original question model to obtain the target question model.

[0029] In some embodiments, the sample question data is obtained, including:

[0030] Obtaining training question data;

[0031] Performing semantic recognition on the training question data to obtain a training semantic vector;

[0032] Performing semantic matching on the training semantic vector and a preset semantic vector to obtain a second matching result;

[0033] Taking the training semantic vector matched by the second matching result as a key semantic vector, the sample question data is screened from the training question data according to the key semantic vector.

[0034] To achieve the above-mentioned purpose, a second aspect of the embodiments of the present application proposes a question and answer method, the method comprising:

[0035] Obtaining to-be-answered data;

[0036] Performing semantic matching on the to-be-answered data and target question and answer data in a preset question and answer corpus to obtain a third matching result; wherein the question and answer corpus is constructed according to the method of the first aspect;

[0037] Taking the target question and answer data matched by the third matching result as key question data; wherein the key question and answer data includes key reply data;

[0038] Performing reply processing according to the key reply data to obtain answer data of the to-be-answered data.

[0039] To achieve the above-mentioned purpose, a third aspect of the embodiments of the present application proposes a question and answer corpus construction device, the device comprising:

[0040] A data acquisition module is configured to acquire historical dialogue data; wherein the historical dialogue data includes historical question and answer data and dialogue identification data; wherein the dialogue identification data is used to identify question and answer information of the historical question and answer data;

[0041] a data segmentation module, configured to perform data segmentation processing on the historical question and answer data according to the dialogue identification data, to obtain initial question and answer data;

[0042] a first data screening module, configured to perform data amount screening on the initial question and answer data according to a first preset data amount, to obtain candidate question and answer data; wherein the candidate question and answer data comprises original question data;

[0043] a content recognition module, configured to perform content recognition on the original question data according to a preset target question model, to obtain a target question label of the original question data; wherein the target question label is used to indicate that a content category of the original question data is a question category or a non-question category;

[0044] a second data screening module, configured to perform screening on the candidate question and answer data according to the target question label, to obtain target question and answer data;

[0045] a question and answer corpus construction module, configured to construct a question and answer corpus according to the target question and answer data.

[0046] To achieve the above object, a fourth aspect of the embodiments of the present application provides an electronic device, the electronic device comprising a memory and a processor, the memory storing a computer program, and the processor implementing the method of the first aspect or the second aspect when executing the computer program.

[0047] To achieve the above object, a fifth aspect of the embodiments of the present application provides a computer readable storage medium, the computer readable storage medium storing a computer program, and the computer program implementing the method of the first aspect or the second aspect when executed by a processor.

[0048] The question-and-answer corpus construction method, apparatus, device, and medium proposed in this application perform content recognition on the original question data through a target question model to obtain corresponding target question tags. Candidate question-and-answer data is then filtered using these target question tags to obtain target question-and-answer data, which is then used to construct the question-and-answer corpus. Therefore, this application avoids the method of constructing a question-and-answer corpus through manual annotation in related technologies. This application improves the efficiency of question-and-answer corpus construction by using data output from the target question model. Furthermore, before performing content recognition based on the target question model, this application also performs data segmentation on historical question-and-answer data based on dialogue identifier data and filters the initial question-and-answer data based on a first preset data volume. This reduces the amount of data input into the target question model while ensuring the quality of the input data, thus improving the efficiency of question-and-answer corpus construction and ensuring its accuracy. When this application is applied to insurance scenarios in fintech, it can improve the efficiency and accuracy of constructing a question-and-answer corpus for insurance scenarios. Attached Figure Description

[0049] Figure 1 This is a flowchart of the method for constructing a question-and-answer corpus provided in the embodiments of this application;

[0050] Figure 2 yes Figure 1 The flowchart of step S103 in the process;

[0051] Figure 3 yes Figure 2 The flowchart of step S204 in the process;

[0052] Figure 4 yes Figure 1 The flowchart also includes steps S104 and above;

[0053] Figure 5 yes Figure 4 The flowchart of step S401 in the text;

[0054] Figure 6 yes Figure 1 The flowchart of step S105 in the process;

[0055] Figure 7 This is a flowchart of the question-and-answer method provided in the embodiments of this application;

[0056] Figure 8 This is a schematic diagram of the structure of the question-and-answer corpus construction apparatus provided in the embodiments of this application;

[0057] Figure 9 This is a schematic diagram of the hardware structure of the electronic device provided in the embodiments of this application. DETAILED DESCRIPTION

[0058] In order to make the purposes, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and not intended to limit the present application.

[0059] It should be noted that although the functional modules are divided in the device schematic diagram, and the logical order is shown in the flowchart, in some cases, the steps shown or described can be performed in a manner different from the module division in the device or the order in the flowchart. The terms "first", "second", etc. in the specification and claims and the above-described drawings are used to distinguish similar objects, and do not necessarily describe a specific order or sequence.

[0060] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of the present application and are not intended to limit the present application.

[0061] First, the terms involved in the present application are analyzed:

[0062] Artificial intelligence (AI): is a new technical science to study, develop the theory, method, technology and application system for simulating, extending and expanding human intelligence; artificial intelligence is a branch of computer science, artificial intelligence tries to understand the essence of intelligence, and produces a new intelligent machine that can react in a similar way to human intelligence, the research in this field includes robots, language recognition, image recognition, natural language processing and expert system, etc. Artificial intelligence can simulate the information process of human consciousness and thinking. Artificial intelligence is also the theory, method, technology and application system of using digital computer or digital computer controlled machine to simulate, extend and expand human intelligence, to perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0063] Natural language processing (NLP): NLP uses computers to process, understand and use human languages (such as Chinese, English, etc.), and NLP is a branch of artificial intelligence, an interdisciplinary subject of computer science and linguistics, and is also commonly referred to as computational linguistics. Natural language processing includes syntax analysis, semantic analysis, discourse understanding, etc. Natural language processing is commonly used in machine translation, handwritten and printed character recognition, speech recognition and text-to-speech conversion, information intent recognition, information extraction and filtering, text classification and clustering, public opinion analysis and opinion mining, etc. It involves data mining, machine learning, knowledge acquisition, knowledge engineering, artificial intelligence research related to language processing, and linguistic research related to language computing.

[0064] At present, in the field of financial technology, intelligent answers can be given to user questions based on a question and answer corpus, for example, intelligent answers can be given to user questions about insurance consulting. In related technologies, a question and answer corpus is constructed or the question and answer corpus is expanded by manual annotation. However, the above method needs to consume a long manual annotation time, thereby affecting the construction efficiency of the question and answer corpus. Therefore, how to improve the construction efficiency of the question and answer corpus has become a technical problem to be solved.

[0065] Based on this, the embodiments of the present application provide a question and answer corpus construction method and device, a question and answer method, equipment and medium, aiming to improve the construction efficiency of the question and answer corpus.

[0066] The question and answer corpus construction method and device, the question and answer method, the equipment and the medium provided by the embodiments of the present application are specifically explained by the following embodiments, and first, the recommendation method in the embodiments of the present application is described.

[0067] The embodiments of the present application can acquire and process related data based on artificial intelligence technology. Among them, artificial intelligence (AI) is to use digital computers or digital computer controlled machines to simulate, extend and expand human intelligence, perceive environment, acquire knowledge and use knowledge to obtain the best results.

[0068] The basic technology of artificial intelligence generally includes technologies such as sensors, special artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction system, mechatronics, etc. The software technology of artificial intelligence mainly includes computer vision technology, robot technology, biometric technology, speech processing technology, natural language processing technology and machine learning / deep learning, etc.

[0069] The question and answer corpus construction method provided in the embodiments of the present application relates to the technical field of financial technology. The question and answer corpus construction method provided in the embodiments of the present application can be applied to a terminal, can also be applied to a server end, and can also be software running in the terminal or the server end. In some embodiments, the terminal can be a smart phone, a tablet computer, a notebook computer, a desktop computer, etc.; the server end can be configured as a stand-alone physical server, can also be configured as a server cluster or a distributed system formed by multiple physical servers, and can also be configured as a cloud server providing basic cloud computing services such as cloud service, cloud database, cloud computing, cloud function, cloud storage, network service, cloud communication, middleware service, domain name service, security service, CDN, and big data and artificial intelligence platform; and the software can be an application for implementing the question and answer corpus construction method, etc., but is not limited to the above forms.

[0070] The present application can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc. The present application can be described in the general context of computer-executable instructions executed by a computer, such as a program module. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present application can also be practiced in a distributed computing environment, in which tasks are performed by remote processing devices connected by a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media, including storage devices.

[0071] It should be noted that in each specific embodiment of the present application, when relevant processing needs to be performed on data related to the identity or characteristics of the user, such as user information, user dialogue data, user behavior data, user history data, and user location information, the user's permission or consent will be obtained first, and the collection, use, and processing of these data will comply with relevant laws, regulations, and standards. In addition, when the embodiments of the present application need to obtain sensitive personal information of the user, the separate permission or separate consent of the user will be obtained through a pop-up window or a jump to a confirmation page, and after obtaining the separate permission or separate consent of the user, the necessary user-related data for enabling the embodiments of the present application to normally operate will be obtained.

[0072] Figure 1 is an optional flowchart of the question and answer corpus construction method provided by the embodiments of the present application, Figure 1 The method in can include but is not limited to steps S101 to S106.

[0073] In step S101, historical dialogue data is acquired; wherein the historical dialogue data comprises historical question and answer data and dialogue identification data; wherein the dialogue identification data is used to identify question and answer information of the historical question and answer data.

[0074] In step S102, data segmentation processing is performed on the historical question and answer data according to the dialogue identification data, to obtain initial question and answer data.

[0075] In step S103, data volume screening is performed on the initial question and answer data according to a first preset data volume, to obtain candidate question and answer data; wherein the candidate question and answer data comprises original question data.

[0076] In step S104, content recognition is performed on the original question data according to a preset target question model, to obtain a target question label of the original question data; wherein the target question label is used to indicate that a content category of the original question data is a question category or a non-question category.

[0077] In step S105, screening is performed on the candidate question and answer data according to the target question label, to obtain target question and answer data.

[0078] In step S106, a question and answer corpus is constructed according to the target question and answer data.

[0079] The steps S101 to S106 shown in the embodiments of the present application, through the target question model, content recognition is performed on the original question data, to obtain a corresponding target question label. Through the target question label, screening is performed on the candidate question and answer data, to obtain target question and answer data, and then a question and answer corpus is constructed according to the target question and answer data. Therefore, it can be known that the embodiments of the present application avoid the method of constructing a question and answer corpus by manual labeling in the related art, and the question and answer corpus constructed by the data output by the target question model can improve the construction efficiency of the question and answer corpus. Moreover, before content recognition is performed according to the target question model, data segmentation processing is performed on the historical question and answer data according to the dialogue identification data, and data volume screening is performed on the initial question and answer data according to the first preset data volume, so as to reduce the data volume of the data input into the target question model and ensure the data quality of the data input into the target question model, thereby improving the construction efficiency of the question and answer corpus and ensuring the accuracy of the construction of the question and answer corpus. When the present application is applied to the insurance scene in the financial technology, the construction efficiency and the construction accuracy of the question and answer corpus of the insurance scene can be improved.

[0080] In step S101 of some embodiments, the historical dialogue data of the questioner and the respondent is obtained through an Application Programming Interface (API) or the like. It can be understood that the historical dialogue data includes historical question and answer data and dialogue identification data. The historical question and answer data includes question data of the questioner and answer data generated by the respondent when answering the question data. The historical question and answer data is a collection of multiple question data and multiple answer data, i.e., the historical question and answer data can include question data of multiple questioners. For example, when the present application is applied to a terminal, the terminal can be installed with an application having an answering function, such as an e-commerce application, a voice assistant, a blog forum, etc. The question data corresponding to different questioners and the corresponding answer data are obtained according to the API interface of the above-mentioned application, thereby constructing the historical question and answer data. When the present application is applied to an insurance scenario, the historical question and answer data can include question data such as “What risks are covered by this insurance? What is the premium?” and answer data such as “This insurance covers risks such as accidental death, hospitalization allowance, etc., and the premium is XX yuan per year”. When an application includes multiple business functions, in order to improve the efficiency and accuracy of subsequent answering processing according to the question and answer corpus, different historical question and answer data can be constructed according to different business functions. For example, the terminal is installed with an insurance management application, which includes insurance consultation function, insurance purchase function, etc. At this time, the historical question and answer data corresponding to the insurance consultation function and the historical question and answer data corresponding to the insurance purchase function are obtained according to the above-mentioned method, respectively.

[0081] In addition, the data category of the historical question and answer data can be any one of text, image, and audio. For example, for the insurance scenario, the question data is obtained through the text data input by the questioner in the customer service interface of the insurance management application; for the voice assistant, the question data is obtained through the audio data of the questioner collected by the audio collection device of the terminal. It can be understood that in order to facilitate data analysis, when the historical question and answer data includes question data of different data categories and / or answer data of different data categories, the data categories need to be normalized. Specifically, the data of the image category is converted to the data of the text category through Optical Charater Recognition (OCR), and the data of the audio category is converted to the data of the text category through a voice recognition method.

[0082] It can be understood that, since the historical question and answer data is a collection of multiple question data and multiple answer data, it can be known that the historical question and answer data has a large amount of data. Different question data can be generated by different question objects, or by the same question object at different times. Therefore, in order to be able to data split the large amount of historical question and answer data, so as to facilitate subsequent data analysis, the historical dialogue data further includes multiple dialogue identification data. The dialogue identification data is used to identify the question and answer information of the question data and the answer data corresponding to the question data, wherein the question and answer information includes time information, login information, object information, etc. For example, when the question and answer information includes time information, the dialogue identification data is the generation time of the question data and the answer data corresponding to the question data. For example, the dialogue identification data of question data A1 is February 2, 2023, 9:00, the answer data corresponding to the question data A1 includes data A2 and data A3, the dialogue identification data of data A2 is February 2, 2023, 9:01, and the dialogue identification data of data A3 is February 2, 2023, 9:02. When the question and answer information includes login information, the dialogue identification data is the session ID. It can be understood that, when the terminal receives the question request instruction, such as when the terminal receives the question request instruction generated by the question object through the sending control of the customer service interface, the terminal will allocate a session ID for the question object. In this session, the question data generated by the question object and the answer data generated by the answer object according to the question data will have the same session ID. Wherein, the current session refers to the session before the terminal receives the question end instruction, and the question end instruction can be generated by the question object touching the interface closing control of the customer service interface. When the question and answer information includes object information, since different question objects have different login IDs in the application, the login ID can also be used as the dialogue identification data.

[0083] In step S102 of some embodiments, since the historical question and answer data has a large amount of data, the historical question and answer data should also be subjected to data segmentation processing to obtain initial question and answer data, so that in subsequent steps, the initial question and answer data with a smaller data amount is used as a data analysis unit, thereby providing analysis efficiency and analysis accuracy of data analysis, and further improving the construction efficiency and construction accuracy of the question and answer corpus. Specifically, the historical question and answer data is subjected to data segmentation according to the dialogue identification data to obtain a plurality of initial question and answer data. For example, when the dialogue identification data is time data, the question data and the reply data within a preset time length are segmented into one initial question and answer data. For example, when the preset time length is two minutes, the question data A1, data A2 and data A3 are segmented into one initial question and answer data. When the dialogue identification data is a session ID, the question data and the reply data with the same session ID are segmented into one initial question and answer data. When the dialogue identification data is a login ID, the question data and the reply data with the same login ID are segmented into one initial question and answer data. It can be understood that according to actual needs, the historical question and answer data can also be subjected to data segmentation processing according to the combination of different dialogue identification data, for example, the dialogue identification data of time data and the dialogue identification data of login ID are combined, and the embodiments of the present application are not limited in this regard.

[0084] It can be understood that since the generation of the session ID depends on the question end instruction, and the question object usually controls the interface close control when the question data is completely replied. Therefore, compared with using time data and login ID as dialogue identification data, using session ID as dialogue identification data can improve the integrity of the initial question and answer data, that is, the reply data in the initial question and answer data can reply the question data more comprehensively. In order to facilitate the description, the session ID is taken as an example of dialogue identification data in the following embodiments.

[0085] In step S103 of some embodiments, the initial question and answer data includes question data and reply data of one session, and therefore in order to avoid the phenomenon that the data amount of the initial question and answer data is still large due to the question object generating more question data in one session, the initial question and answer data needs to be subjected to data amount screening according to a first preset data amount. Specifically, the initial question and answer data with a data amount less than or equal to the first preset data amount is taken as candidate question and answer data, and the initial question and answer data with a data amount greater than the first preset data amount is filtered out.

[0086] Referring to Figure 2 In some embodiments, step S103 includes but is not limited to steps S201 to S204.

[0087] In step S201, the initial question and answer data is subjected to semantic recognition to obtain an initial semantic vector.

[0088] Step S202, performing semantic matching between the initial semantic vector and the preset semantic vector to obtain a first matching result.

[0089] Step S203, taking the initial semantic vector matched by the first matching result as a target semantic vector, and filtering out the target semantic data from the initial question data to obtain the key question and answer data.

[0090] Step S204, performing data volume screening on the key question and answer data according to a first preset data volume to obtain candidate question and answer data.

[0091] In step S201 of some embodiments, the initial question and answer data is subjected to semantic recognition by NLP or the like to obtain an initial semantic vector corresponding to the question data and an initial semantic vector corresponding to the reply data in the initial question and answer data.

[0092] In step S202 of some embodiments, the preset semantic vector is set in advance according to the data content of the expected candidate question and answer data. For example, the expected candidate question and answer data is data that does not include invalid dialogue such as greetings, and the preset semantic vector corresponding to the greetings such as "hello" and "delighted to serve you" is set. The preset semantic vector is subjected to semantic matching with the initial semantic vector to obtain a first matching result. The first matching result is used to represent whether the initial semantic vector is a semantic vector corresponding to the greetings, i.e., whether the question data or the reply data corresponding to the initial semantic vector is a greeting.

[0093] In step S203 of some embodiments, when the first matching result indicates semantic matching, it indicates that the semantics of the initial semantic vector is content such as greetings. At this time, the initial semantic vector is taken as a target semantic vector, and the question data or the reply data corresponding to the target semantic vector is taken as target semantic data. The target semantic data in the initial question data is filtered out, and the remaining data in the filtered initial question data is taken as the key question and answer data, thereby realizing preliminary data cleaning of the question data and / or the reply data in the initial question data.

[0094] In step S204 of some embodiments, the data volume of the key question and answer data after preliminary data cleaning may still be large. Therefore, the key question and answer data is screened again according to a first preset data volume to obtain candidate question and answer data. Specifically, the key question and answer data with a data volume less than or equal to the first preset data volume is taken as the candidate question and answer data, and the key question and answer data with a data volume greater than the first preset data volume is filtered out.

[0095] Reference Figure 3 In some embodiments, the key question and answer data includes key question data, and step S204 includes but is not limited to steps S301 to S302.

[0096] In step S301, the question data of the key question data is obtained, and the amount of question data is obtained.

[0097] In step S302, if the amount of question data is less than the first preset data amount, the key question and answer data is taken as the candidate question and answer data.

[0098] It can be understood that, since the data amount of the reply data in the key question and answer data is generally associated with the data amount of the question data, that is, the more question data generated by the question object, the more reply data corresponding to the reply object needs to generate. Therefore, by restricting the data amount of the question data in the key question and answer data, the purpose of restricting the overall data amount of the candidate question and answer data can be achieved. Different from the above embodiment, the above embodiment restricts the overall data amount of the key question and answer data by the first preset data amount, while the embodiment of the application restricts the question data in the key question and answer data by the first preset data amount. Therefore, the embodiment of the application not only can ensure that the overall data amount of the candidate question and answer data obtained by screening is small, but also can ensure that the data amount of the question data in the candidate question and answer data obtained by screening is controllable.

[0099] In step S301 of some embodiments, the question data in the key question and answer data is taken as the key question data, the data amount of the key question data in the key question and answer data is determined, and the amount of question data is obtained.

[0100] In step S302 of some embodiments, the amount of question data is compared with the first preset data amount. Assuming that the value of the first preset data amount is 3, if the amount of question data is less than the first preset data amount, it indicates that the corresponding key question data contains less key question data. Less key question data is beneficial to question content semantic analysis and reply content semantic analysis. Therefore, the key question and answer data with the amount of question data less than the first preset data amount is taken as the candidate question and answer data.

[0101] Reference Figure 4 In some embodiments, before step S104, the method provided by the embodiment of the application further includes training a target question model, specifically including but not limited to steps S401 to S403.

[0102] In step S401, sample question data and sample question data sample question labels are obtained; wherein the sample question label is used to represent that the content category of the sample question data is a question category or a non-question category;

[0103] In step S402, the content of the sample question data is identified according to a preset original question model, and an original question label is obtained; wherein the original question label is used to represent that the content category of the sample question data is a question category or a non-question category;

[0104] In step S403, the original question model is adjusted in parameters according to the sample question label and the original question label, to obtain a target question model.

[0105] In step S401 of some embodiments, sample question data and sample question labels of the sample question data obtained by labeling or the like are acquired. The sample question label is used to represent that the content category of the sample question data is a question category or a non-question category. It can be understood that the acquisition method of the sample question data can refer to the acquisition method of the historical question and answer data, and the embodiments of the present application will not be repeated. Similarly, the data category of the sample question data can be any one of text, image, and audio, and the embodiments of the present application will not be specifically limited. It should be noted that since the sample question data is only used to train the target question model, and the training target of the target question model is to accurately identify the content category of the sample question data, that is, the training of the target question model is irrelevant to the reply data. Therefore, unlike the question data in the historical question and answer data, the question data in the historical question and answer data must have a mapping relationship with a certain reply data, and the sample question data can contain or not contain such a mapping relationship, that is, the sample question data can be data only with a question without a corresponding reply. When the semantic content represented by the reply data is used to answer the semantic content represented by the question data, it is determined that the question data and the reply data have a mapping relationship.

[0106] Reference Figure 5 In some embodiments, step S401 includes but is not limited to steps S501 to S504.

[0107] In step S501, training question data is acquired.

[0108] In step S502, semantic recognition is performed on the training question data to obtain a training semantic vector.

[0109] In step S503, the training semantic vector is matched with a preset semantic vector to obtain a second matching result.

[0110] In step S504, the training semantic vector matched in the second matching result is taken as a key semantic vector, and the sample question data is selected from the training question data according to the key semantic vector.

[0111] In step S501 of some embodiments, the training question data is acquired by a related API interface or the like. It can be understood that the training question data is data generated by a question object, and the specific acquisition method of the training question data can refer to the acquisition method of the historical question data, and the embodiments of the present application will not be repeated.

[0112] In step S502 of some embodiments, semantic recognition is performed on the training question data by NLP or the like to obtain a training semantic vector corresponding to the training question data.

[0113] In step S503 of some embodiments, a preset semantic vector corresponding to the greeting data such as greetings is preset, for example, a preset semantic vector corresponding to the greeting data such as "hello" and "thank you for your answer". The preset semantic vector is matched with the training semantic vector to obtain a second matching result. The second matching result is used to represent whether the training semantic vector is a semantic vector corresponding to the greeting data, that is, whether the training question data corresponding to the training semantic vector is the greeting data.

[0114] In step S504 of some embodiments, when the second matching result indicates semantic matching, it indicates that the semantic of the training semantic vector is the content such as greetings. At this time, the training semantic vector is taken as a key semantic vector, and the training question data corresponding to the key semantic vector is taken as key semantic data. The key semantic data is filtered, and the remaining training question data after filtering is taken as sample question data, so as to realize data cleaning of the training question data.

[0115] In step S402 of some embodiments, an original question model based on a classifier structure or a transform structure is constructed in advance, so the model structure of the original question model can be any one of the following: Bidirectional Encoder Representation from Transformer (BERT) derived from transform, support vector machine, logistic regression, decision tree, etc. The sample question data is taken as input data of the original question model, so that the content of the sample question data is recognized by the original question model, and the sample question data is determined to be question data or non-question data according to the recognition result, so as to obtain an original question label corresponding to the sample question data. If the original question label represents that the content category of the sample question data is a question category, it indicates that the sample question data is question data. If the original question label represents that the content category of the sample question data is a non-question category, it indicates that the sample question data is non-question data. The question data refers to data that can accurately determine the question content, and the non-question data refers to data that cannot obtain the question content. For example, "I want to consult the guarantee range of XX insurance" is question data, and "my name is Xiaoming, and I am 20 years old" is non-question data.

[0116] In step S403 of some embodiments, the sample question label and the original question label are loss calculated according to a preset loss function to determine the content recognition error of the original question model. The original question model is parameter adjusted according to the calculated loss value to obtain a target question model with more accurate content recognition.

[0117] In step S104 of some embodiments, the question data in the screened candidate question and answer data is taken as original question data. A target question model is pre-trained to accurately identify data content to determine the content category. The original question data is taken as input data of the target question model to identify the content of the original question data by the target question model, and the content category of the original question data is determined according to the identification result, so as to obtain the target question label corresponding to the original question data.

[0118] In step S105 of some embodiments, the target question label is used to represent that the content category of the original question data in the candidate question and answer data is a question category or a non-question category, so the candidate question and answer data containing the original question data can also be screened according to the target question label to screen the desired target question and answer data. The desired target question and answer data can be candidate question and answer data whose content category of the included original question data is all question category, or candidate question and answer data whose proportion of the original question data of the question category to the original question data of the non-question category is a preset proportion, which is not limited in the embodiments of the present application.

[0119] Reference Figure 6 In some embodiments, the target question label includes a question positive label, and the question positive label is used to represent that the content category of the original question data is a question category. Step S105 includes but is not limited to steps S601 to S603.

[0120] Step S601, screening target question data from the original question data according to the question positive label;

[0121] Step S602, obtaining the data amount of the target question data in the candidate question and answer data to obtain a target data amount;

[0122] Step S603, if the target data amount is less than a second preset data amount, taking the candidate question and answer data as target question and answer data.

[0123] In step S601 of some embodiments, the target question label includes a question positive label and a question negative label. The question positive label is used to represent that the content category of the original question data is a question category, and the question negative label is used to represent that the content category of the original question data is a non-question category. The original question data with the question positive label is taken as the target question data according to the question positive label.

[0124] In step S602 of some embodiments, the data amount of the target question data in the candidate question data is obtained, that is, the data amount of the question data generated by the question object in the candidate question and answer data is determined to obtain the target data amount.

[0125] In step S603 of some embodiments, a second preset data amount with a small value is set, for example, the value of the second preset data amount is 2. If the target data amount is less than the second preset data amount, it indicates that the corresponding candidate question and answer data only includes one target question data. At this time, all the reply data in the candidate question and answer data is data for answering the target question data. Therefore, the candidate question and answer data with a target data amount less than the second preset data amount is taken as the target question and answer data, avoiding the need to re-divide the data in content to determine the mapping relationship between the reply data and the question data in the candidate question and answer data, that is, to determine which reply data is used to answer one question data in the candidate question and answer data and which reply data is used to answer another question data in the candidate question and answer data.

[0126] In step S106 of some embodiments, the target question and answer data obtained through screening is integrated to construct a question and answer corpus. Specifically, the plurality of target question and answer data can also be clustered to cluster the target question and answer data with similar semantics of original question data into one question and answer corpus, for example, a question and answer corpus with semantics of original question data for consulting B insurance and a question and answer corpus with semantics of original question data for consulting C insurance can be clustered. Thus, the processing efficiency of subsequent reply processing of the to-be-answered data according to the question and answer corpus can be improved.

[0127] The method for constructing a question and answer corpus provided by the embodiments of the present application performs content recognition on the original question data through a target question model to obtain a corresponding target question label. The target question and answer data is obtained by screening the candidate question and answer data through the target question label, and then a question and answer corpus is constructed according to the target question and answer data. It can be seen that the embodiments of the present application avoid the method of constructing a question and answer corpus by manual labeling in related technologies, and the data output by the target question model is used to construct a question and answer corpus, which can improve the construction efficiency of the question and answer corpus. Moreover, before content recognition according to the target question model, the embodiments of the present application also perform data division processing according to the dialogue identification data and data amount screening on the initial question and answer data according to the first preset data amount, thereby reducing the data amount of the data input into the target question model and ensuring the data quality of the data input into the target question model, so that the construction efficiency of the question and answer corpus can be improved, and the accuracy of the construction of the question and answer corpus can also be ensured. When the present application is applied to the insurance scene in the financial technology, the construction efficiency and construction accuracy of the question and answer corpus in the insurance scene can be improved.

[0128] Reference Figure 7 The embodiments of the present application also provide a question and answer method, which includes but is not limited to steps S701 to S704.

[0129] Step S701, acquiring to-be-answered data;

[0130] In step S702, the to-be-answered data is semantically matched with the target question and answer data in the preset question and answer corpus to obtain a third matching result; wherein the question and answer corpus is obtained according to the construction method of the question and answer corpus;

[0131] In step S703, the target question and answer data matched in the third matching result is taken as the key question data; wherein the key question and answer data includes key reply data.

[0132] In step S704, the answer data of the to-be-answered data is obtained by reply processing according to the key reply data.

[0133] In step S701 of some embodiments, the to-be-answered data is obtained through a related API interface. It can be understood that the to-be-answered data is the to-be-answered question data generated by the question object, and the obtaining method of the to-be-answered data can refer to the obtaining method of the historical question and answer data, which will not be repeated here. Similarly, the data category of the to-be-answered data can also be any one of text, image and audio, which will not be limited here.

[0134] In step S702 of some embodiments, the to-be-answered data is semantically matched with the target question and answer data obtained according to the question and answer corpus construction method described in any of the above embodiments, and the corresponding third matching result is obtained. It can be understood that when the constructed question and answer corpus includes target question and answer data with multiple semantic contents, such as target question and answer data with semantic content of insurance consultation, target question and answer data with semantic content of insurance purchase, target question and answer data with semantic content of train ticket information query, target question and answer data with semantic content of airplane ticket information query, etc., the to-be-answered data needs to be semantically matched with each target question and answer data in the question and answer corpus to determine the semantic content of the to-be-answered data. In addition, when a plurality of question and answer corpora are constructed by clustering a plurality of target question and answer data, a library tag can be constructed for each question and answer corpus, which is used to represent the semantic content of all target question and answer data in the question and answer corpus. For example, a first question and answer corpus with a library tag of insurance and a second question and answer corpus with a library tag of train ticket are constructed. At this time, the to-be-answered data only needs to be semantically matched with the library tag, thereby improving the semantic matching efficiency.

[0135] In steps S703 to S704 of some embodiments, the third matching result representing the semantic matching target question and answer data is taken as key question and answer data, and the answer data in the key question and answer data is taken as key answer data. According to the key answer data and the actual application scenario, corresponding answer processing is performed, for example, when the data to be answered is data generated by the user in the terminal application interface in an input manner, the key answer data is taken as answer data, and the answer data is displayed in the same application interface. When the data to be answered is data obtained by the voice assistant, the key answer data is converted into audio data, and the converted audio data is taken as answer data.

[0136] The embodiments of the present application can improve the accuracy of the answer processing by using the question and answer corpus described in the above embodiments, because the question and answer corpus constructed according to the above embodiments has high accuracy.

[0137] Reference Figure 8 The embodiments of the present application also provide a question and answer corpus construction device, which can implement the question and answer corpus construction method described above. The device comprises:

[0138] The data acquisition module 801 is configured to acquire historical dialogue data. The historical dialogue data comprises historical question and answer data and dialogue identification data. The dialogue identification data is used to identify the question and answer information of the historical question and answer data.

[0139] The data segmentation module 802 is configured to perform data segmentation processing on the historical question and answer data according to the dialogue identification data, to obtain initial question and answer data.

[0140] The first data screening module 803 is configured to perform data volume screening on the initial question and answer data according to a first preset data volume, to obtain candidate question and answer data. The candidate question and answer data comprises original question data.

[0141] The content recognition module 804 is configured to perform content recognition on the original question data according to a preset target question model, to obtain a target question label of the original question data. The target question label is used to indicate that the content category of the original question data is a question category or a non-question category.

[0142] The second data screening module 805 is configured to screen the candidate question and answer data according to the target question label, to obtain target question and answer data.

[0143] The question and answer corpus construction module 806 is configured to construct a question and answer corpus according to the target question and answer data.

[0144] The specific implementation of the question and answer corpus construction device is basically the same as that of the above-described question and answer corpus construction method, and will not be described here again.

[0145] The embodiment of the present application further provides an electronic device, which comprises a memory and a processor, the memory stores a computer program, and the processor implements the above-mentioned method for constructing a question-answer corpus when executing the computer program. The electronic device can be any intelligent terminal including a tablet computer, a vehicle-mounted computer and the like.

[0146] Referring to Figure 9 , Figure 9 The hardware structure of the electronic device of another embodiment is illustrated, which comprises:

[0147] The processor 901 can be implemented in the form of a general-purpose CPU (Central Processing Unit), a microprocessor, an ASIC (Application Specific Integrated Circuit), or one or more integrated circuits, and is used to execute related programs to implement the technical solutions provided by the embodiments of the present application.

[0148] The memory 902 can be implemented in the form of a ROM (ReadOnly Memory), a static storage device, a dynamic storage device or a RAM (Random Access Memory). The memory 902 can store an operating system and other application programs, and when the technical solutions provided by the embodiments of the present application are implemented by software or firmware, the related program codes are stored in the memory 902 and called and executed by the processor 901 to implement the method for constructing a question-answer corpus according to the embodiments of the present application.

[0149] The input / output interface 903 is used to realize information input and output.

[0150] The communication interface 904 is used to realize the communication interaction between the device and other devices, and can realize communication through a wired manner (for example, a USB, a network cable and the like) or a wireless manner (for example, a mobile network, WIFI, Bluetooth and the like).

[0151] The bus 905 transmits information between various components (for example, the processor 901, the memory 902, the input / output interface 903 and the communication interface 904) of the device.

[0152] The processor 901, the memory 902, the input / output interface 903 and the communication interface 904 are connected to each other through the bus 905 to realize communication connection between them in the device.

[0153] The embodiment of the present application further provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the above-mentioned method for constructing a question-answer corpus.

[0154] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. In addition, the memory can include a high-speed random access memory and can also include a non-transitory memory, such as at least one magnetic disk storage device, a flash memory device, or other non-transitory solid-state memory device. In some embodiments, the memory can optionally include a memory that is remotely disposed relative to the processor, and these remote memories can be connected to the processor through a network. Examples of the above network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0155] The embodiments described in the embodiments of the present application are for more clearly illustrating the technical solutions of the embodiments of the present application, and do not constitute a limitation on the technical solutions provided by the embodiments of the present application. Those skilled in the art can know that, with the evolution of technology and the appearance of new application scenarios, the technical solutions provided by the embodiments of the present application are also applicable to similar technical problems.

[0156] Those skilled in the art can understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of the present application, and can include more or fewer steps than shown in the figures, or combine certain steps, or different steps.

[0157] The device embodiments described above are only schematic, and the units described as separate components can or can not be physically separate, i.e., can be located in one place, or can be distributed on multiple network units. Part or all of the modules can be selected according to actual needs to achieve the purpose of the embodiments of the present application.

[0158] Those skilled in the art can understand that all or some of the steps in the above disclosed method, the functional modules / units in the system and the device can be implemented as software, firmware, hardware and their appropriate combinations.

[0159] The terms "first", "second", "third", "fourth" and the like used in the specification of the present application and the above-described drawings, if any, are used to distinguish similar objects, and do not necessarily have to be used to describe a particular order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device including a series of steps or units does not have to be limited to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0160] It should be understood that, in the application, "at least one" refers to one or more, and "multiple" refers to two or more. "And / or" is used to describe the association relationship of the associated objects, which means that there can be three relationships, for example, "A and / or B" can represent three cases of only A, only B, and A and B existing at the same time, wherein A and B can be singular or plural. The character " / " generally represents an "or" relationship between the associated objects before and after it. "At least one of the following" or similar expressions means any combination of these items, including any combination of single or multiple items. For example, at least one of a, b or c can represent a, b, c, "a and b", "a and c", "b and c", or "a and b and c", wherein a, b, and c can be single or multiple.

[0161] In several embodiments provided in the application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of the above units is only a logical function division, and actual implementation can have another division manner, for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the displayed or discussed mutual ones can be indirect coupling or communication connection through some interfaces, devices or units, and can be electrical, mechanical or other forms.

[0162] The units described above as separate components can or can not be physically separated, and the components shown as units can or can not be physical units, that is, they can be located in one place, or can be distributed on multiple network units. According to actual needs, part or all of the units can be selected to achieve the purpose of the embodiment scheme.

[0163] In addition, each functional unit in each embodiment of the application can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of a software functional unit.

[0164] The integrated unit, if implemented in the form of a software function unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application, essentially or in other words, the part that contributes to the prior art or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes multiple instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various program storage media.

[0165] The preferred embodiments of the embodiments of the present application are described above with reference to the accompanying drawings, and are not limited to the scope of the embodiments of the present application. Any modifications, equivalent replacements and improvements made by those skilled in the art without departing from the scope and essence of the embodiments of the present application shall be within the scope of the embodiments of the present application.

Claims

1. A method for constructing a question-answer corpus, characterized by, The method comprises: acquiring historical dialogue data; wherein the historical dialogue data comprises historical question and answer data and dialogue identification data; wherein the dialogue identification data is used to identify question and answer information of the historical question and answer data; performing data segmentation processing on the historical question and answer data according to the dialogue identification data, to obtain initial question and answer data; performing data amount screening on the initial question and answer data according to a first preset data amount, to obtain candidate question and answer data; wherein the candidate question and answer data comprises original question data; performing content recognition on the original question data according to a preset target question model, to obtain a target question label of the original question data; wherein the target question label is used to indicate that the content category of the original question data is a question category or a non-question category; screening the candidate question and answer data according to the target question label, to obtain target question and answer data; constructing a question and answer corpus according to the target question and answer data; wherein the target question label comprises a question positive label, and the question positive label is used to indicate that the content category of the original question data is a question category; the screening of the candidate question and answer data according to the target question label to obtain target question and answer data comprises: screening target question data from the original question data according to the question positive label; obtaining a target data amount of the target question data in the candidate question and answer data; if the target data amount is less than a second preset data amount, regarding the candidate question and answer data as the target question and answer data; before the content recognition on the original question data according to the preset target question model to obtain the target question label of the original question data, the method further comprises training the target question model, specifically comprising: obtaining sample question data and a sample question label of the sample question data; wherein the sample question label is used to represent that the content category of the sample question data is a question category or a non-question category; performing content recognition on the sample question data according to a preset original question model, to obtain an original question label; wherein the original question label is used to represent that the data category of the sample question data is a question category or a non-question category; performing parameter adjustment on the original question model according to the sample question label and the original question label, to obtain the target question model.

2. The method of claim 1, wherein, the data amount screening of the initial question and answer data according to the first preset data amount to obtain the candidate question and answer data comprises: performing semantic recognition on the initial question and answer data, to obtain an initial semantic vector; performing semantic matching on the initial semantic vector and a preset semantic vector, to obtain a first matching result; regarding the initial semantic vector matched by the first matching result as a target semantic vector, and screening key question and answer data from the initial question and answer data according to the target semantic vector; performing data amount screening on the key question and answer data according to the first preset data amount, to obtain the candidate question and answer data.

3. The method of claim 2, wherein, the key question and answer data comprises key question data; the data amount screening of the key question and answer data according to the first preset data amount to obtain the candidate question and answer data comprises: obtaining question data of the key question data, to obtain a quantity of question data; if the quantity of question data is less than the first preset quantity of data, taking the key question and answer data as the candidate question and answer data.

4. The method of claim 1, wherein, The method comprises: obtaining training question data; performing semantic recognition on the training question data to obtain a training semantic vector; performing semantic matching on the training semantic vector and a preset semantic vector to obtain a second matching result; taking the training semantic vector that represents semantic matching as a key semantic vector, and screening the sample question data from the training question data according to the key semantic vector.

5. A question and answer method characterized by, The method comprises: obtaining to-be-answered data; performing semantic matching on the to-be-answered data and target question and answer data in a preset question and answer corpus to obtain a third matching result; wherein the question and answer corpus is obtained by the method according to any one of claims 1 to 4; taking the target question and answer data that represents semantic matching as key question and answer data; wherein the key question and answer data comprises key reply data; performing reply processing according to the key reply data to obtain answer data of the to-be-answered data.

6. An apparatus for constructing a question-answer corpus, characterized by comprising: The device comprises: a data acquisition module configured to acquire historical dialogue data; wherein the historical dialogue data comprises historical question and answer data and dialogue identification data; wherein the dialogue identification data is used to identify question and answer information of the historical question and answer data; a data segmentation module configured to perform data segmentation processing on the historical question and answer data according to the dialogue identification data to obtain initial question and answer data; a first data screening module configured to perform data quantity screening on the initial question and answer data according to a first preset quantity of data to obtain candidate question and answer data; wherein the candidate question and answer data comprises original question data; a content recognition module configured to perform content recognition on the original question data according to a preset target question model to obtain a target question label of the original question data; wherein the target question label is used to represent that the content category of the original question data is a question category or a non-question category; a second data screening module configured to screen the candidate question and answer data according to the target question label to obtain target question and answer data; a question and answer corpus construction module configured to construct a question and answer corpus according to the target question and answer data.

7. An electronic device, comprising: The electronic device comprises a memory and a processor, the memory stores a computer program, and the processor implements the method according to any one of claims 1 to 4 or the method according to claim 5 when executing the computer program.

8. A computer-readable storage medium storing a computer program, the computer program comprising instructions that, when executed by a computer, cause the computer to perform the method of any one of claims 1 to 7. The computer program is executed by the processor to implement the method according to any one of claims 1 to 4 or the method according to claim 5.

Images