Model training method, data processing method, electronic device and storage medium

Abstract

The present application discloses a model training method, a data processing method, an electronic device and a storage medium, and relates to the fields of large model technology and computer technology. The method comprises: obtaining a target training data set, wherein the target training data set is obtained by expanding the original training data set, and the original training data set comprises: multiple sets of training data, and the training data comprises: original questions, original search fragment sequences and original answers; training based on the target training data set to obtain a target query processing model, wherein the target query processing model is used to query the currently input query question to obtain a query reply. The present application solves the technical problems of poor flexibility of the query processing method provided in the related art and low accuracy of the query results.

Description

Technical Field

[0001] The present application relates to the fields of large model technology and computer technology, and specifically, to a model training method, a data processing method, an electronic device and a storage medium. Background Art

[0002] With the rise of big model technology, more and more artificial intelligence (AI) assistants are able to understand users' intentions and needs through natural language processing and machine learning technologies, thereby providing more accurate and personalized services. Taking AI assistants in audio and video content scenarios as an example, by quickly refining and precipitating knowledge through big model technology, users can efficiently complete the transcription, retrieval, summary and organization of audio and video content. When users ask customized questions based on the uploaded audio and video content, the query processing results obtained will be unreliable and less flexible due to the non-standard questioning methods and derivative questions based on the audio and video content.

[0003] To address the above-mentioned problems, no effective solution has been proposed yet. Summary of the invention

[0004] The embodiments of the present application provide a model training method, a data processing method, an electronic device and a storage medium to at least solve the technical problems of poor flexibility and low accuracy of query results in the query processing method provided in the related art.

[0005] According to one aspect of an embodiment of the present application, a model training method is provided, including: obtaining a target training data set, wherein the target training data set is obtained by expanding an original training data set, the original training data set includes: multiple groups of training data, the training data include: original questions, original search fragment sequences and original answers; training based on the target training data set to obtain a target query processing model, wherein the target query processing model is used to perform query processing on a currently input query question to obtain a query reply.

[0006] According to another aspect of an embodiment of the present application, a data processing method is also provided, including: obtaining a currently input query question; performing query processing on the query question using a target query processing model to obtain a query reply; wherein the target query processing model is trained based on a target training data set, and the target training data set is obtained by expanding an original training data set, and the original training data set includes: multiple groups of training data, and the training data includes: an original question, an original search fragment sequence, and an original answer.

[0007] According to another aspect of an embodiment of the present application, a data processing method is also provided, including: obtaining a currently input video conference query question; using a conference content query processing model to query and process the video conference query question to obtain a video conference query reply; wherein the conference content query processing model is trained based on a target conference content training data set, and the target conference content training data set is obtained by expanding an original conference content training data set, and the original conference content training data set includes: multiple groups of conference content training data, and the conference content training data includes: original conference content questions, original conference content search fragment sequences, and original conference content answers.

[0008] According to another aspect of an embodiment of the present application, a data processing method is also provided, including: obtaining a query processing request through a first application programming interface; returning a query processing response through a second application programming interface; wherein the request data carried in the query processing request includes: a query question, and the response data carried in the query processing response includes: a query reply, and the query reply is obtained by query processing the query question using a target query processing model, and the target query processing model is trained based on a target training data set, and the target training data set is obtained by expanding an original training data set, and the original training data set includes: multiple groups of training data, and the training data includes: an original question, an original search fragment sequence, and an original answer.

[0009] According to another aspect of an embodiment of the present application, a data processing method is also provided, including: obtaining a currently input query processing dialogue request; returning a query processing dialogue reply in response to the query processing dialogue request; wherein the request data carried in the query processing dialogue request includes: a query question, and the information carried in the query processing dialogue reply includes: a query reply, and the query reply is obtained by query processing the query question using a target query processing model, the target query processing model is trained based on a target training data set, the target training data set is obtained by expanding an original training data set, and the original training data set includes: multiple groups of training data, and the training data includes: an original question, an original search fragment sequence, and an original answer; and the query reply is displayed in a graphical user interface.

[0010] According to another aspect of an embodiment of the present application, an electronic device is also provided, including: a memory storing an executable program; and a processor for running the program, wherein when the program is running, any model training method or data processing method of the embodiments of the present application is executed.

[0011] According to another aspect of an embodiment of the present application, a computer-readable storage medium is also provided, and the computer-readable storage medium includes a stored executable program, wherein when the executable program is running, the device where the computer-readable storage medium is located is controlled to execute any one of the model training methods or data processing methods in the embodiments of the present application.

[0012] According to another aspect of the embodiments of the present application, a computer program product is also provided, including a computer program, which implements the model training method or data processing method of any one of the embodiments of the present application when executed by a processor.

[0013] In an embodiment of the present application, a target training data set is obtained by expanding an original training data set, and then a target query processing model is obtained by training based on the target training data set. The target query processing model can be used to perform query processing on the currently input query question to obtain a query reply. The accuracy and universality of the query question are judged by using the target query processing model, and then the wrong query question is corrected before answering, or a richer answer other than the original text is provided under appropriate circumstances, thereby providing more robust and richer answers, helping users to obtain more information input, and achieving the purpose of quickly obtaining accurate query replies to the currently input query questions, thereby achieving the technical effect of improving the flexibility of query processing and the accuracy of query results, and thus solving the technical problems of poor flexibility of the query processing method provided in the related art and low accuracy of the query results.

[0014] It is easy to notice that the above general description and the following detailed description are only for the purpose of exemplifying and explaining the present application, and do not constitute a limitation of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0016] Figure 1 This is a schematic diagram of an application scenario of a model training method according to Example 1 of the present application;

[0017] Figure 2 is a flow chart of a model training method according to Embodiment 1 of the present application;

[0018] Figure 3A is a schematic diagram of an expansion process of an original search fragment sequence according to Example 1 of the present application;

[0019] Figure 3B is a schematic diagram of another expansion process of the original search fragment sequence according to Example 1 of the present application;

[0020] Figure 4 is a schematic diagram of a data set expansion process according to Example 1 of the present application;

[0021] Figure 5 is a flow chart of a data processing method according to Embodiment 2 of the present application;

[0022] Figure 6 is a flow chart of a data processing method according to Embodiment 3 of the present application;

[0023] Figure 7 is a flow chart of a data processing method according to Embodiment 4 of the present application;

[0024] Figure 8 is a flow chart of a data processing method according to Embodiment 5 of the present application;

[0025] Fig. 9 is a structural block diagram of a model training device according to Embodiment 6 of the present application;

[0026] Fig.10 is a structural block diagram of a data processing device according to Embodiment 6 of the present application;

[0027] Fig.11 is a structural block diagram of another data processing device according to Embodiment 6 of the present application;

[0028] Fig.12 is a structural block diagram of another data processing device according to Embodiment 6 of the present application;

[0029] Fig.13 is a structural block diagram of another data processing device according to Embodiment 6 of the present application;

[0030] Fig.14 It is a structural block diagram of a computer terminal according to Example 7 of the present application. DETAILED DESCRIPTION

[0031] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0032] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0033] The technical solution provided in this application is mainly implemented using large-scale model technology. The large model here refers to a deep learning model with large-scale model parameters, which can usually contain hundreds of millions, tens of billions, hundreds of billions, trillions or even more than 10 trillion model parameters. The large model can also be called a foundation model / foundation model. The large model is pre-trained with large-scale unlabeled corpus to produce a pre-trained model with more than 100 million parameters. This model can adapt to a wide range of downstream tasks, and the model has good generalization ability, such as large-scale language model (Large Language Model, LLM), multi-modal pre-training model, etc.

[0034] It should be noted that when the large model is actually applied, the pre-trained model can be fine-tuned through a small number of samples so that the large model can be applied to different tasks. For example, the large model can be widely used in natural language processing (NLP), computer vision, speech processing and other fields, and can be specifically applied to computer vision tasks such as visual question answering (VQA), image caption (IC), image generation, etc. It can also be widely used in natural language processing tasks such as text-based sentiment classification, text summary generation, and machine translation. Therefore, the main application scenarios of the large model include but are not limited to digital assistants, intelligent robots, search, online education, office software, e-commerce, intelligent design, etc. In the embodiment of the present application, data processing through a conference content query processing model in a video conferencing scenario is used as an example for explanation.

[0035] First, some nouns or terms that appear in the description of the embodiments of the present application are subject to the following explanations:

[0036] Retrieval-Augmented Generation (RAG) system: is an artificial intelligence mechanism that combines retrieval and generation to improve the performance of language models in information retrieval and knowledge-intensive tasks. First, the retrieval component retrieves information or document fragments related to the input query from a large knowledge base or dataset, and then uses these retrieved contents to assist the generation component to produce more accurate and information-rich output.

[0037] Search chunk: refers to the smallest unit of retrieval in RAG, usually a chunk of 300 to 500 words. The combination of multiple most relevant chunks will serve as contextual information for RAG input when answering questions.

[0038] In the related technology, intelligent question-and-answer assistant products used in audio and video scenarios support users to ask customized questions about uploaded audio records, and conduct quick questions and searches in a more natural human-computer interaction method. However, in actual use, free input also brings the following problems: (1) The way of expressing questions varies, and it is not a standardized way of asking questions. There may be unclear intentions or even a small number of errors; (2) The user's intention to ask questions is clear, but it is a derivative question of the input content, which means that the uploaded audio and video does not contain the answer to the question. Faced with these non-standard questions that do not necessarily have correct answers, simply refusing to answer or directly providing a wrong answer will greatly affect the user experience.

[0039] One of the related technologies is a large-model-based cloud disk smart assistant that can understand multimodal information in typical scenarios such as work, life, and study, so as to achieve fast file and video search, knowledge point summary, document translation, and even content creation. Users can ask questions and get answers based on the audio and video resources in the user's cloud disk, but the cloud disk smart assistant does not answer entirely based on the video content. It will also search for specific questions and introduce encyclopedia knowledge to give richer and more complete answers. The disadvantage of this method is that users cannot distinguish whether the answer comes from the original text or from external knowledge.

[0040] One type of conference intelligent assistant in the related art is presented in the form of a question-and-answer box. The user can check the lower left corner of the question-and-answer box to determine whether to answer based on the uploaded recording file, so that the user can actively control the source of the required answer. However, this method is less flexible, and the user needs to choose whether to answer based on the recording file, so it is impossible to get both the answer based on the original text and the general answer based on external knowledge at the same time.

[0041] Example 1

[0042] According to an embodiment of the present application, a query processing method is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0043] Considering the huge amount of model parameters of the large model and the limited computing resources of the mobile terminal, the above model training method provided in the embodiment of the present application can be applied to Figure 1 The application scenarios shown are not limited to these. Figure 1 In the application scenario shown, the large model is deployed in the server 10, and the server 10 can be connected to one or more client devices 20 through a local area network connection, a wide area network connection, an Internet connection, or other types of data networks. The client device 20 here may include but is not limited to: a smart phone, a tablet computer, a laptop computer, a PDA, a personal computer, a smart home device, a vehicle-mounted device, etc. The client device 20 can interact with the user through a graphical user interface to implement the call of the large model, thereby implementing the method provided in the embodiment of the present application.

[0044] In an embodiment of the present application, a system composed of a client device and a server can perform the following steps: a target training data set can be obtained in the server, wherein the target training data set is obtained by expanding the original training data set, and the original training data set includes: multiple sets of training data, and the training data includes: original questions, original search fragment sequences and original answers, and then the target query processing model is obtained by training based on the target training data set. Then the server obtains the currently input query processing dialogue request from the client device, and the server responds to the query processing dialogue request and returns a query processing dialogue reply to the client device; wherein the request data carried in the query processing dialogue request includes: query questions, and the information carried in the query processing dialogue reply includes: query replies, and the query replies are obtained by query processing the query questions using the target query processing model; finally, the query replies are displayed in the graphical user interface of the client device. It should be noted that the embodiment of the present application can be performed in the client device when the operating resources of the client device can meet the deployment and operation conditions of the large model.

[0045] Under the above operating environment, this application provides Figure 2 The model training method shown. Figure 2 1 is a flow chart of a model training method according to Example 1 of the present application. Figure 2 As shown, the method may include the following steps:

[0046] Step S21, obtaining a target training data set, wherein the target training data set is obtained by expanding an original training data set, and the original training data set includes: multiple sets of training data, and the training data includes: an original question, an original search fragment sequence, and an original answer;

[0047] Step S22, training a target query processing model based on the target training data set, wherein the target query processing model is used to perform query processing on the currently input query question to obtain a query reply.

[0048] The model training method in the embodiment of the present application can be applied to audio and video content scenarios with high knowledge added value, such as meetings, classes, interviews, training, interviews, live broadcasts, and watching videos.

[0049] The above-mentioned original training data set D includes: multiple groups of training data, each training data is defined as d={q, [s1, s2, s3], a}, where q represents the original question, [s1, s2, s3] represents the original search fragment sequence, and a represents the original answer. The original question is a specific question or query raised by the user, and the original is the basis of the training data set. The original question can be open-ended or closed, and is used to define the content that the target query processing model needs to understand and answer. The original search fragment sequence is an information fragment related to the original question, which can come from different sources, such as web pages, documents, databases, etc. The search fragment sequence is the original material used by the target query processing model to find the original answer. The original answer is the correct answer or response corresponding to the original question. During the training process, the target query processing model needs to learn how to extract or generate the original answer from the original search fragment sequence.

[0050] For example, in the process of obtaining the original training data set, the original questions and related answers are first collected from different sources, which may include web searches, question-and-answer forums, professional databases, etc. Then, irrelevant or low-quality data are removed to ensure the accuracy and reliability of the original training data set. For original questions that do not have ready-made answers, the answers can be manually annotated, or semi-supervised learning methods can be used to assist.

[0051] Expanding the original training data set to obtain the target training data set can improve the performance and generalization ability of the target query processing model. After training the target query processing model based on the target training data set, the target query processing model is used to query the current input query question to obtain a query response.

[0052] Based on the above steps S21 to S22, a target training data set is obtained by expanding the original training data set, and then a target query processing model is obtained based on the target training data set. The target query processing model can be used to perform query processing on the currently input query question to obtain a query reply. The accuracy and universality of the query question are judged by using the target query processing model, and then the wrong query question is corrected before answering, or a richer answer other than the original text is provided under appropriate circumstances, thereby providing more robust and richer answers, helping users to obtain more information input, and achieving the purpose of quickly obtaining accurate query replies to the currently input query question, thereby achieving the technical effect of improving the flexibility of query processing and the accuracy of query results, and thus solving the technical problems of poor flexibility of the query processing method provided in the related technology and low accuracy of the query results.

[0053] The model training method in the embodiment of the present application is further introduced below.

[0054] In an optional embodiment, in step S21, obtaining a target training data set includes:

[0055] Step S211, filtering the original training data set according to a preset filtering condition to obtain a filtered training data set, wherein the preset filtering condition is set based on the matching degree between the original question and the corresponding original answer;

[0056] Step S212, expanding the filtered training data set to obtain a target training data set.

[0057] Specifically, a function f(x, y)={0, 1} based on string mapping is defined to determine the relationship between a given question x and an answer y. For a given question x, if the answer y is an informative answer, the value of f(x, y) is 1, otherwise the value of f(x, y) is 0, for example, because there is no answer or recognition is refused for security reasons. In order to further improve the richness of the answers, the original training data set is filtered according to the preset filtering conditions to obtain a filtered training data set. The preset filtering conditions are set based on the degree of match between the original question and the corresponding original answer. If the length of the original answer in the original training data set is too short or the original answer cannot provide information, the degree of match between the original question and the original answer is low.

[0058] The above preset screening conditions may include: the original answer length len(s) is less than a preset length threshold. For example, the preset length threshold may be set to 100, that is, the training samples whose original answer length is less than 100 need to be further expanded to further enrich the training samples; f(q, a) = 0, that is, for the original question q input, the original answer a cannot give an informative answer.

[0059] Based on the above optional embodiments, the original training data set is filtered according to preset screening conditions to obtain a filtered training data set, and then the filtered training data set is expanded to obtain a target training data set, thereby further improving the data quality of the target training data set, so as to further improve the query processing effect of the target query processing model.

[0060] In an optional embodiment, in step S212, the filtered training data set is expanded to obtain a target training data set including:

[0061] Step S2121, expanding the query content of the original question in the filtered training data set to obtain the target question;

[0062] Step S2122, expanding the original search segment sequence in the filtered training data set to obtain a target search segment sequence;

[0063] Step S2123, expanding the original answer based on the target question and the target search segment sequence to obtain a target answer;

[0064] Step S2124, obtaining a target training data set based on the target question, the target search fragment sequence and the target answer.

[0065] Specifically, the query content of the original question in the filtered training data set is expanded, including but not limited to rewriting, error correction and other processing processes, thereby obtaining the target question, which can be the correct question obtained after rewriting and error correction. The original search fragment sequence in the filtered training data set is expanded with retrieval fragments to obtain a target search fragment sequence. Based on the target question and the target search fragment sequence, the original answer is expanded to obtain the target answer, and finally the target training data set is obtained based on the target question, the target search fragment sequence and the target answer.

[0066] Based on the above optional embodiment, by expanding the query content of the original question in the filtered training data set, the target question is obtained, and then the original search fragment sequence in the filtered training data set is expanded to obtain the target search fragment sequence, and then the original answer is expanded based on the target question and the target search fragment sequence to obtain the target answer, and finally the target training data set is obtained based on the target question, the target search fragment sequence and the target answer, which can more comprehensively cover the query question currently input, thereby improving the generalization ability of the training target query processing model. At the same time, by expanding the original answer, the target query processing model can improve its ability to understand and generate answers of different forms, further improving the accuracy of the query processing results.

[0067] In an optional embodiment, in step S2121, the query content of the original question in the filtered training data set is expanded to obtain the target question, including: obtaining a first prompt text, wherein the first prompt text is used to prompt to correct the query content of the original question based on a given document, or, rewrite the query content of the original question based on the given document and correct the rewritten question; based on the first prompt text, the query content of the original question in the filtered training data set is expanded to obtain the target question.

[0068] Specifically, the first prompt text is used to indicate a question reformulation prompt. First, determine whether the original question contains entity errors, such as errors in names of people, places, times, etc. If the original question contains entity errors, correct the query content of the original question based on the given document, and continue answering after obtaining the target question. If the original question does not contain entity errors, rewrite the query content of the original question based on the given document and correct the rewritten question. Specifically, the original question can be reformed into a flawed question with a smaller probability, and then correct the errors. After obtaining the target question, continue answering.

[0069] Exemplarily, the content of the first prompt text may be: Given the following multiple document knowledge and corresponding questions, please output the results according to the following instructions: The return format is a lightweight data exchange format (json), including whether the query content can be corrected (if_correct), the original question (origin_query), the rewritten question (rewrite_query), and the reason why the rewritten question is wrong (correct_reason).

[0070] The requirement for question rewriting is: replace entities of this type in the original question with names of people, places, and times that have appeared in the document, and keep the semantics of the question itself smooth. For example, in a given document, A and B are chatting, and A makes a suggestion. The original question is what A's suggestion is, which can be rewritten as what B's suggestion is. If the original question cannot recognize the name of a person, place, or time for rewriting, if_correct is 0, and rewrite_query and correct_reason are both empty strings. Otherwise, if_correct is 1, and the remaining fields need to be completed. If the original question can be rewritten successfully, the reason for the error needs to be supplemented, that is, to explain which part of the question after rewriting is problematic and inconsistent with the content mentioned in the given document.

[0071] For example, the original text mentioned that Li and Wang were chatting, and that Wang's knee was injured because of playing basketball. The original question was: "What was the reason for Wang's knee injury?". By transforming the prompt output by the above question, the json is: {"if_correct": 1, "origin_query": "What was the reason for Wang's knee injury?", "rewrite_query": "What was the reason for Li's knee injury?", "correct_reason": "This question may be wrong. The person mentioned in the original text of the given document who had a knee injury was Wang, not Li"}.

[0072] Based on the above optional embodiments, by obtaining the first prompt text, and then expanding the query content of the original question in the filtered training data set based on the first prompt text, the target question is obtained, thereby improving the accuracy and semantic understanding of the query content of the original question, and at the same time expanding the query questions in the target training data set, thereby improving the generalization ability of the target query processing model and its adaptability to diversified queries.

[0073] In an optional embodiment, in step S2122, the original search fragment sequence in the filtered training data set is expanded to obtain a target search fragment sequence, including: in response to the original search fragment sequence containing multiple consecutive search fragments extracted from the same document, based on the position information of the original search fragment sequence in the same document and a preset search fragment expansion number, the number of search fragments in the original search fragment sequence in the filtered training data set is expanded to obtain a target search fragment sequence.

[0074] Specifically, when the original search fragment sequence contains multiple continuous search fragments extracted from the same document, that is, for samples in which the original input document is a continuous chunk in a single document scenario, the number of search fragments in the original search fragment sequence in the filtered training data set is expanded based on the position information of the original search fragment sequence in the same document and the preset number of search fragment expansions. While ensuring the continuity of the current chunk, the position of the chunk to be expanded is first determined. Specifically, any of the following expansions can be performed on the basis of the current chunk: forward expansion, backward expansion, and simultaneous expansion of the front and back, so as to introduce as much information as possible.

[0075] Figure 3A is a schematic diagram of an expansion process of an original search fragment sequence according to Example 1 of the present application, such as Figure 3AAs shown, the original search fragment sequence contains multiple search fragments extracted from the same document: search fragment k, search fragment i, search fragment j, which are sorted to obtain multiple continuous search fragments: search fragment i, search fragment j, search fragment k. Based on the position information of the original search fragment sequence in the same document and the preset search fragment expansion number 2, the number of search fragments of the original search fragment sequence in the filtered training data set is expanded. Case 1 is the target search fragment sequence obtained by expanding 2 search fragments forward: search fragment i-2, search fragment i-1, search fragment i, search fragment j, search fragment k. Case 2 is the target search fragment sequence obtained by expanding 1 search fragment forward and backward at the same time: search fragment i-1, search fragment i, search fragment j, search fragment k, search fragment k+1. Case 3 is the target search fragment sequence obtained by expanding 2 search fragments backward: search fragment i, search fragment j, search fragment k, search fragment k+1, search fragment k+2.

[0076] Based on the above optional embodiments, in response to the original search fragment sequence containing multiple continuous search fragments extracted from the same document, based on the position information of the original search fragment sequence in the same document and the preset search fragment expansion number, the number of search fragments of the original search fragment sequence in the filtered training data set is expanded to obtain a target search fragment sequence, thereby improving the diversity and coverage of the query results, thereby more accurately meeting the needs of users.

[0077] In an optional embodiment, in step S2122, the original search fragment sequence in the filtered training data set is expanded to obtain a target search fragment sequence, including: in response to the original search fragment sequence containing multiple search fragments extracted from multiple documents, retrieving a preset number of search fragments from the target document library according to the original search fragment sequence; while retaining the original search fragment sequence, based on the comparison results between the preset number of search fragments and the original search fragment sequence and the preset search fragment expansion number, the number of search fragments in the original search fragment sequence in the filtered training data set is expanded to obtain the target search fragment sequence.

[0078] Specifically, in a cross-document scenario, when the original search fragment sequence contains multiple search fragments extracted from multiple documents, the top 10 most relevant search fragments are first retrieved from the entire target document library according to the original question corresponding to the original search fragment sequence, that is, the top10chunk is obtained, and then compared with the original search fragment sequence one by one in descending order of relevance. If the fragment is not covered, a new fragment is added. In this way, the original chunk is retained while the two most relevant chunks are supplemented.

[0079] Figure 3Bis a schematic diagram of another expansion process of the original search fragment sequence according to Example 1 of the present application, such as Figure 3B As shown, the original search fragment sequence includes multiple search fragments extracted from multiple documents: search fragment a, search fragment b, search fragment c, and the first 10 search fragments are retrieved from the target document library based on the original search fragment sequence. Under the premise of retaining the original search fragment sequence, the number of search fragments in the original search fragment sequence in the filtered training data set is expanded according to the comparison results between the preset number of search fragments and the original search fragment sequence and the preset search fragment expansion number 10, and the target search fragment sequence is obtained: search fragment ranked 1 (a), search fragment ranked 2, search fragment ranked 3, search fragment ranked 4 (b), and search fragment ranked 6 (c).

[0080] Based on the above optional embodiments, in response to the original search fragment sequence containing multiple search fragments extracted from multiple documents respectively, a preset number of search fragments are retrieved from the target document library according to the original search fragment sequence, and then on the premise of retaining the original search fragment sequence, based on the comparison results between the preset number of search fragments and the original search fragment sequence and the preset search fragment expansion number, the number of search fragments of the original search fragment sequence in the filtered training data set is expanded to obtain the target search fragment sequence, thereby improving the diversity and coverage of the query results, thereby more accurately meeting the needs of users.

[0081] In an optional embodiment, the target question includes: a first part of question content and a second part of question content. In step S2123, the original answer is expanded based on the target question and the target search fragment sequence to obtain the target answer, including: obtaining a first answer corresponding to the first part of the question content based on general knowledge, and obtaining a second answer corresponding to the second part of the question content based on the target search fragment sequence; by fusing the first answer and the second answer, the original answer is expanded to obtain the target answer.

[0082] The first part of the question content is the question content that can be answered without given reference information, that is, general questions, such as common sense questions. The second part of the question content is the question content that can only be answered based on given reference information, that is, specific questions, and the subject of the inquiry is a specific person or thing, for example, the conclusion of a meeting, the opinion of a certain anchor. The first answer corresponding to the first part of the question content can be obtained based on the general world knowledge, and the second answer corresponding to the second part of the question content can be obtained based on the target search fragment sequence. Finally, by fusing the first answer and the second answer, the original answer is expanded to obtain the target answer. The above-mentioned general knowledge can be common knowledge, which specifically involves various fields, such as history, geography, science, culture, etc., and specifically refers to knowledge that is widely disseminated and accepted in daily life, and usually does not require specific professional background or training to understand and apply. General knowledge can be mainly used to understand the world, make decisions and solve problems. When answering query questions, using general knowledge can quickly find the first answer and reduce query search time.

[0083] Based on the above optional embodiments, a first answer corresponding to the first part of the question content is obtained based on general knowledge, and a second answer corresponding to the second part of the question content is obtained based on the target search fragment sequence, and finally the original answer is expanded by fusing the first answer and the second answer to obtain the target answer, thereby improving the completeness and accuracy of the target answer, and further improving the query flexibility and robustness of the target query processing model.

[0084] In an optional embodiment, obtaining a first answer corresponding to a first part of question content based on general knowledge includes: obtaining a second prompt text, wherein the second prompt text is used to prompt whether the target question depends on a given document and provide a corresponding answer for a first part of question content in the target question that does not depend on the given document; based on the second prompt text, obtaining a first answer corresponding to the first part of question content from general knowledge.

[0085] The second prompt text mentioned above is the prompt for the general question judgment step, which determines whether the input target question q is a general question or a specific question. At the same time, it requires the reason for the judgment and the first answer. The reason for the judgment can assist the model in judging the universality of the question, and the first answer can further supplement the general world knowledge.

[0086] For example, the prompt of the general question judgment step may be: Given a question, the specific task is to determine whether relevant documents must be provided for this question to be answered, and try to provide an answer without reference as much as possible, and return an output in json format. The specific requirements are:

[0087] The returned json format contains three keys: the first is type, which indicates whether the question must have relevant documents to answer; the second is reason, please give the reason for your judgment, that is, why the question must or must not rely on relevant documents; the third is answer, that is, provide an answer as much as possible. The value range of type is {0,1}, 0 means that the question is a relatively general and common sense question that can be answered without referring to documents, and 1 means that the question involves specific people or events, and reference documents must be given to accurately answer it. Reason represents the basis for judgment, such as the original text mentioning a person or a meeting, which cannot be covered by general knowledge. Answer represents the answer given. If type is 0, please give a general answer directly, and be as detailed as possible; if type is 1, you can also try to answer a relatively general part of the question, or guess the possible answer with some basis.

[0088] For example, suppose the input question is "In the traffic promotion system, how do merchants obtain traffic through tasks and activities?", the output result obtained through the above prompt is: {"type": 0, "reason": "The question asks about the general way for merchants to obtain traffic in the traffic promotion system. This is part of the promotion knowledge and does not require specific document reference", "answer": "In the traffic promotion system, merchants usually obtain traffic by setting various tasks to attract customer participation and organizing promotional activities. Tasks may include interactive forms such as signing in, sharing to social platforms, and inviting new users. These tasks can stimulate user participation and provide them with certain rewards, such as discount cards, points or small gifts, thereby prompting users to return and increase the probability of purchase. Activities may include limited-time discounts, group purchases, flash sales, etc. These activities stimulate consumers to make purchasing decisions as soon as possible by providing short-term discounts, and also increase the exposure of merchants. Merchants also often attract potential customers through search engine optimization and search engine promotion, social media promotion, content promotion, etc. Overall, the goal is to create an actively involved community environment and increase brand awareness, which will be converted into actual traffic and sales"}.

[0089] Based on the above optional embodiment, by obtaining the second prompt text, it is possible to clearly determine whether the target question depends on the given document, and to provide a corresponding answer for the first part of the question content in the target question that does not depend on the given document. Based on the second prompt text, the first answer corresponding to the first part of the question content can be obtained from general knowledge, thereby improving the accuracy and comprehensiveness of the question answer.

[0090] In an optional embodiment, obtaining a second answer corresponding to the second part of the question content based on the target search fragment sequence includes: obtaining a third prompt text, wherein the third prompt text is used to prompt retrieval of a given document and generation of an answer corresponding to the second part of the question content; based on the third prompt text, obtaining the second answer corresponding to the second part of the question content from the target search fragment sequence.

[0091] Specifically, the third prompt text is the prompt of the two-way answer generation step. The two-way in the two-way answer generation refers to giving answers "based only on the question" or "given documents and questions", where the first-way answer is obtained when the universality of the question is judged. The second-way answer is based on the given document and question, which is a standard RAG process, that is, s=f(q, S), where q and s are the question and answer respectively, and S is the expanded target search segment sequence [s1, s2, s3, s4, s5].

[0092] For example, in order to get more accurate answers and citation sources, the following is a prompt for constructing training data: Given multiple text fragments arranged in chronological order and indexed by [id], please answer the question based on the given text and return an output in json format. The specific requirements are:

[0093] The returned JSON format contains 4 keys: the first is type, which indicates whether the question can be answered; the second is answer, which is the answer to the question; the third is query, which is the question itself; and the fourth is index, which is the source of the original text. The value range of type is {-1, 0, 1}, where -1 indicates that the answer to the question is not mentioned in the original text, 0 indicates that part of the question can be answered based on the original text, and 1 indicates that the answer to the question is completely mentioned in the original text. If type is -1, please indicate in the answer why the question has no answer in the original text; if type is 0, while giving the correct answer to the answerable part, you need to point out the unanswerable part of the question; if type is 1, just give the correct answer. The content of the answer must comply with the content and expression of the original text. You can summarize, introduce relevant background and explanations required for the answer, and enrich the details of the answer, but you cannot generate details that do not exist in the original text. The format of index is the order of fragment indexes, for example [1][2], which means that the answer comes from fragments [1] and [2]. Note that the index cannot exceed the fragment range of the original text.

[0094] For example, suppose the input question is "In the traffic promotion system, how do merchants obtain traffic through tasks and activities?", the output example is as follows: {"query": "In the traffic promotion system, how do merchants obtain traffic through tasks and activities?", "type": 0, "answer": "Merchants can guide users to participate by publishing related tasks and some activities, such as grabbing red envelopes, installing applications, playing games, etc. to obtain traffic, but the specific ways to obtain traffic through these activities are not detailed in the original text", "index": "[5]"}.

[0095] Based on the above optional embodiment, by obtaining the third prompt text, and then based on the third prompt text, obtaining the second answer corresponding to the second part of the question content from the target search fragment sequence, thereby improving the accuracy and comprehensiveness of the answer to the question.

[0096] In an optional embodiment, the target answer is obtained by fusing the first answer and the second answer to expand the original answer, which includes: obtaining a fourth prompt text, wherein the fourth prompt text is used to prompt the distinction of sources of answers to different parts of the target question content; fusing the first answer and the second answer based on the fourth prompt text to obtain a fusion result; and expanding the original answer based on the fusion result to obtain the target answer.

[0097] The fourth prompt text is the prompt of the answer fusion step. In the end, it is necessary to make a choice and merge the first answer and the second answer, so as to retain the original answer as much as possible. At the same time, if the question has a certain degree of universality, appropriate world knowledge can be supplemented. Table 1 shows the labels of whether the reference document can be answered according to the judgment of the general question, and details the fusion strategy provided by the embodiment of this application:

[0098] Table 1 Fusion strategy

[0099]

[0100] Specifically, when the RAG answer is 1, that is, when the question is completely answerable, for general questions, RAG knowledge is fully trusted, and general knowledge can be used to simply supplement to get the target answer; for specific questions, only RAG knowledge is retained as the target answer. When the RAG answer is 0, that is, when the question is partially answerable, RAG knowledge has a partial answer to the general question, and general knowledge can be used to reasonably supplement to get the target answer; for specific questions, only RAG knowledge is retained as the target answer. When the RAG answer is -1, that is, when the question is completely unanswerable, RAG knowledge has no answer for both general and specific questions, and general knowledge needs to be used for necessary supplementation to get the target answer.

[0101] For example, when merging, the user needs to be informed of the specific source of the answer, so the fourth prompt text requires a clear distinction between answers from different sources. The prompt for the answer fusion step is as follows:

[0102] Given a question, the specific task is to integrate the answers from two sources, that is, the given reference original text and the non-reference original text, and the final answer is required to be fluent, logically clear, and meaningful. The specific requirements are:

[0103] The final answer needs to have a clear hierarchical structure, first introducing the answer to the given original text (xxx), and then introducing the answer without reference to the original text (yyy), for example: "According to the content of the original text, xxx. Generally speaking, yyy." If the two answers are almost semantically the same, only the answer to the given original text is retained. In principle, the answer to the given original text takes precedence, unless the answer is obviously inconsistent with the question. The answer without reference to the original text can be used as a supplement to the general answer to this question. Pay attention to maintaining the coherence of the final answer. The modal particles, coherent words and restrictive conditions of the two original answers can be deleted. For example, "Since there is no reference document" or "empty", the overlapping parts can be removed from the answer without reference document. If the final answer contains multiple parallel points, the non-standard expression of the list needs to be rewritten in the format of lightweight markup language (markdown) and output uniformly.

[0104] Based on the above optional embodiments, by distinguishing the sources of answers to different parts of the question content, richer information can be obtained, and by integrating answers from different sources, a more comprehensive understanding and solution can be obtained, thereby improving the credibility and accuracy of the target answer, thereby improving user satisfaction with the question query response.

[0105] Figure 4 is a schematic diagram of a data set expansion process according to Embodiment 1 of the present application, Figure 4As shown, the original training data set is filtered according to the preset screening conditions to obtain the filtered training data set, and then the query content of the original question in the filtered training data set is expanded based on the first prompt text to obtain the target question. The first prompt text is used to prompt the query content of the original question to be corrected according to the given document, or the query content of the original question is rewritten according to the given document and the rewritten question is corrected. At the same time, the original search fragment sequence in the filtered training data set is expanded to obtain the target search fragment sequence, which can be specifically supplemented by single document and multiple documents. Further, the first answer corresponding to the general question is obtained based on the general knowledge, and the second answer corresponding to the specific question is obtained based on the target search fragment sequence. Finally, the original answer is expanded by fusing the first answer and the second answer to obtain the target answer. Based on the target question, the target search fragment sequence and the target answer, the target training data set is obtained, and the target query processing model is obtained based on the training of the target training data set.

[0106] The target query processing model obtained by the model training method in the embodiment of the present application can respond to queries freely input by users, by screening short answers and candidate data of the rejection recognition category, and making targeted corrections and constructions to the fault tolerance of the questions, while performing chunk expansion based on continuity and relevance in single-document and multi-document scenarios, and finally, on the answer side, three implementation steps are implemented around general question judgment, two-way answer generation, and label-based answer fusion strategy, which can provide richer answers beyond the original text under appropriate circumstances, thereby providing more robust and richer answers, helping users obtain more information input, and further improving the flexibility of query processing and the accuracy of query results.

[0107] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with the relevant laws, regulations and standards of relevant countries and regions, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0108] It should be noted that, for the aforementioned method embodiments, for the sake of simplicity, they are all expressed as a series of action combinations, but those skilled in the art should be aware that the present application is not limited by the described order of actions, because according to the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present application.

[0109] Through the description of the above implementation methods, those skilled in the art can clearly understand that the method according to the above embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, disk, CD), and includes a number of instructions for a terminal device (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods described in each embodiment of the present application.

[0110] Example 2

[0111] According to an embodiment of the present application, a data processing method is also provided. Figure 5 is a flow chart of a data processing method according to Embodiment 2 of the present application. Figure 5 As shown, the method may include the following steps:

[0112] Step S51, obtaining the currently input query question;

[0113] Step S52, use the target query processing model to query process the query question and obtain a query reply; wherein, the target query processing model is trained based on a target training data set, and the target training data set is obtained by expanding the original training data set, and the original training data set includes: multiple groups of training data, and the training data includes: an original question, an original search fragment sequence and an original answer.

[0114] Based on the above steps S51 to S52, by obtaining the currently input query question, and then using the target query processing model to query process the query question, a query reply is obtained, and the accuracy and universality of the query question are judged by using the target query processing model, and then the wrong query question is corrected before answering, or a richer answer other than the original text is provided under appropriate circumstances, thereby providing more robust and richer answers, helping users to obtain more information input, and achieving the purpose of quickly obtaining accurate query replies to the currently input query questions, thereby achieving the technical effect of improving the flexibility of query processing and the accuracy of query results, and thus solving the technical problems of poor flexibility of the query processing method provided in the relevant technology and low accuracy of the query results.

[0115] The processing object of the above data processing method is multimodal data, which may include but is not limited to: text data, voice data, picture data, video data, etc. As an optional example, the above query question may be a specific question embodied in the form of text data, or the above query question may be a specific question input in the form of voice data, or the above query question may be a specific question presented in the form of picture data / video data.

[0116] In an optional embodiment, a target query processing model is used to perform query processing on a query question to obtain a query response, including: using the target query processing model to retrieve multiple search fragments to be used from multimedia files and / or general knowledge associated with the query question, and generating a query response based on the multiple search fragments to be used.

[0117] For the parts not described in detail in the above embodiments of the present application, please refer to the relevant description of Embodiment 1 and will not be described in detail here.

[0118] Example 3

[0119] According to an embodiment of the present application, a data processing method is also provided. Figure 6 is a flow chart of a data processing method according to Embodiment 3 of the present application. Figure 6 As shown, the method may include the following steps:

[0120] Step S61, obtaining the currently input video conference query question;

[0121] Step S62, use the conference content query processing model to query and process the video conference query question to obtain the video conference query reply; wherein, the conference content query processing model is trained based on the target conference content training data set, and the target conference content training data set is obtained by expanding the original conference content training data set, and the original conference content training data set includes: multiple groups of conference content training data, and the conference content training data includes: original conference content questions, original conference content search fragment sequences and original conference content answers.

[0122] Based on the above steps S61 to S62, by obtaining the currently input video conference query question, and then using the conference content query processing model to query and process the video conference query question, a video conference query reply is obtained. By using the conference content query processing model to judge the accuracy and universality of the query question, the wrong query question is corrected before answering, or a richer answer other than the original text is provided under appropriate circumstances, thereby providing more robust and richer answers, helping users to obtain more information input, and achieving the purpose of quickly obtaining accurate query replies to the currently input query questions, thereby achieving the technical effect of improving the flexibility of query processing and the accuracy of query results, and thus solving the technical problems of poor flexibility of the query processing method provided in the relevant technology and low accuracy of the query results.

[0123] For the parts not described in detail in the above embodiments of the present application, please refer to the relevant description of Embodiment 1 and will not be described in detail here.

[0124] Example 4

[0125] According to an embodiment of the present application, a data processing method is also provided. Figure 7 is a flow chart of a data processing method according to Embodiment 4 of the present application. Figure 7 As shown, the method may include the following steps:

[0126] Step S71, obtaining a query processing request through a first application programming interface;

[0127] Step S72, returning a query processing response through a second application programming interface; wherein the request data carried in the query processing request includes: a query question, and the response data carried in the query processing response includes: a query reply, and the query reply is obtained by query processing the query question using a target query processing model, and the target query processing model is trained based on a target training data set, and the target training data set is obtained by expanding an original training data set, and the original training data set includes: multiple groups of training data, and the training data includes: an original question, an original search fragment sequence, and an original answer.

[0128] Based on the above steps S71 to S72, a query processing request is obtained through the first application programming interface, and then a query processing response is returned through the second application programming interface, thereby achieving the purpose of quickly obtaining an accurate query reply to the currently input query question, thereby achieving the technical effect of improving the flexibility of query processing and the accuracy of query results, and thus solving the technical problems of poor flexibility and low accuracy of query results of the query processing method provided in the relevant technology.

[0129] The first application programming interface and the second application programming interface may be the same application programming interface or different application programming interfaces. In an optional embodiment, the interface parameters in the first application programming interface and the second application programming interface may include but are not limited to: interface global identifier, interface signature key, interface timestamp, interface request identifier, system call credential identifier, etc. The first application programming interface may use GET or POST as the interface request method to obtain the file processing request. The second application programming interface may use JSON format to feedback the file processing response.

[0130] For the parts not described in detail in the above embodiments of the present application, please refer to the relevant description of Embodiment 1 and will not be described in detail here.

[0131] Example 5

[0132] According to an embodiment of the present application, a data processing method is also provided. Figure 8 is a flow chart of a data processing method according to Embodiment 5 of the present application. Figure 8 As shown, the method may include the following steps:

[0133] Step S81, obtaining the currently input query processing dialogue request;

[0134] Step S82, in response to the query processing dialogue request, returning a query processing dialogue reply; wherein the request data carried in the query processing dialogue request includes: a query question, and the information carried in the query processing dialogue reply includes: a query reply, and the query reply is obtained by query processing the query question using a target query processing model, and the target query processing model is trained based on a target training data set, and the target training data set is obtained by expanding an original training data set, and the original training data set includes: multiple sets of training data, and the training data includes: an original question, an original search fragment sequence, and an original answer;

[0135] Step S83, displaying the query response in the graphical user interface.

[0136] Based on the above steps S81 to S83, by obtaining the currently input query processing dialogue request, responding to the query processing dialogue request, returning the query processing dialogue reply, and finally displaying the query reply in the graphical user interface, the purpose of quickly obtaining the accurate query reply of the currently input query question is achieved, thereby achieving the technical effect of improving the flexibility of query processing and the accuracy of query results, and further solving the technical problems of poor flexibility and low accuracy of query results of the query processing method provided in the related technology.

[0137] For the parts not described in detail in the above embodiments of the present application, please refer to the relevant description of Embodiment 1 and will not be described in detail here.

[0138] Example 6

[0139] According to an embodiment of the present application, a model training device for implementing the above-mentioned model training method is also provided. Fig. 9 is a structural block diagram of a model training device according to Example 6 of the present application, such as Fig. 9 As shown, the device comprises:

[0140] An acquisition module 901 is used to acquire a target training data set, wherein the target training data set is obtained by expanding the training data set, and the original training data set includes: multiple sets of training data, and the training data includes: original questions, original search fragment sequences, and original answers;

[0141] The training module 902 is used to train a target query processing model based on a target training data set, wherein the target query processing model is used to perform query processing on a currently input query question to obtain a query reply.

[0142] Optionally, the acquisition module 901 is also used to: filter the original training data set according to preset filtering conditions to obtain a filtered training data set, wherein the preset filtering conditions are set based on the degree of matching between the original question and the corresponding original answer; and expand the filtered training data set to obtain a target training data set.

[0143] Optionally, the acquisition module 901 is also used to: expand the query content of the original question in the filtered training data set to obtain a target question; expand the original search fragment sequence in the filtered training data set to obtain a target search fragment sequence; expand the original answer based on the target question and the target search fragment sequence to obtain a target answer; obtain the target training data set based on the target question, the target search fragment sequence and the target answer.

[0144] Optionally, the acquisition module 901 is also used to: obtain a first prompt text, wherein the first prompt text is used to prompt to correct the query content of the original question based on a given document, or to rewrite the query content of the original question based on a given document and correct the rewritten question; based on the first prompt text, the query content of the original question in the filtered training data set is expanded to obtain the target question.

[0145] Optionally, the acquisition module 901 is also used to: in response to the original search fragment sequence containing multiple consecutive search fragments extracted from the same document, based on the position information of the original search fragment sequence in the same document and a preset search fragment expansion number, expand the number of search fragments of the original search fragment sequence in the filtered training data set to obtain a target search fragment sequence.

[0146] Optionally, the acquisition module 901 is also used to: in response to the original search fragment sequence containing multiple search fragments extracted from multiple documents respectively, retrieve a preset number of search fragments from the target document library according to the original search fragment sequence; on the premise of retaining the original search fragment sequence, expand the number of search fragments in the original search fragment sequence in the filtered training data set according to the comparison results between the preset number of search fragments and the original search fragment sequence and the preset search fragment expansion number, so as to obtain the target search fragment sequence.

[0147] Optionally, the acquisition module 901 is also used to: obtain a first answer corresponding to the first part of the question content based on general knowledge, and obtain a second answer corresponding to the second part of the question content based on the target search fragment sequence; by fusing the first answer and the second answer, the original answer is expanded to obtain the target answer.

[0148] Optionally, the acquisition module 901 is also used to: obtain a second prompt text, wherein the second prompt text is used to prompt whether the target question depends on the given document and provide a corresponding answer for the first part of the question content in the target question that does not depend on the given document; based on the second prompt text, obtain a first answer corresponding to the first part of the question content from general knowledge.

[0149] Optionally, the acquisition module 901 is also used to: obtain a third prompt text, wherein the third prompt text is used to prompt retrieval of a given document and generate an answer corresponding to the second part of the question content; based on the third prompt text, obtain a second answer corresponding to the second part of the question content from the target search fragment sequence.

[0150] Optionally, the acquisition module 901 is also used to: obtain a fourth prompt text, wherein the fourth prompt text is used to prompt the distinction of sources of answers to different parts of the target question; based on the fourth prompt text, the first answer and the second answer are merged to obtain a fusion result; based on the fusion result, the original answer is expanded to obtain the target answer.

[0151] It should be noted that the acquisition module 901 and the training module 902 correspond to steps S21 to S22 in Example 1, and the examples and application scenarios implemented by the two modules and the corresponding steps are the same, but are not limited to the contents disclosed in the above-mentioned Example 1. It should be noted that the above-mentioned modules or units may be hardware components or software components stored in a memory and processed by one or more processors, and the above-mentioned modules may also be run in a computer terminal as part of the device.

[0152] Fig.10 is a structural block diagram of a data processing device according to Embodiment 6 of the present application, such as Fig.10 As shown, the device comprises:

[0153] The acquisition module 1001 is used to acquire the query question currently input;

[0154] The query module 1002 is used to use a target query processing model to query the query question and obtain a query response; wherein the target query processing model is trained based on a target training data set, and the target training data set is obtained by expanding the original training data set, and the original training data set includes: multiple groups of training data, and the training data includes: an original question, an original search fragment sequence and an original answer.

[0155] Optionally, the query module 1002 is further configured to: use a target query processing model to retrieve a plurality of search fragments to be used from multimedia files and / or general knowledge associated with the query question, and generate a query reply based on the plurality of search fragments to be used.

[0156] It should be noted that the acquisition module 1001 and the query module 1002 correspond to steps S51 to S52 in Example 2, and the examples and application scenarios implemented by the two modules and the corresponding steps are the same, but are not limited to the contents disclosed in the above-mentioned Example 2. It should be noted that the above-mentioned modules or units may be hardware components or software components stored in a memory and processed by one or more processors, and the above-mentioned modules may also be run in a computer terminal as part of a device.

[0157] Fig.11 is a structural block diagram of another data processing device according to Embodiment 6 of the present application, such as Fig.11 As shown, the device comprises:

[0158] An acquisition module 1101 is used to acquire a currently input video conference query question;

[0159] The query module 1102 is used to use a conference content query processing model to query and process video conference query questions to obtain video conference query responses; wherein the conference content query processing model is trained based on a target conference content training data set, and the target conference content training data set is obtained by expanding an original conference content training data set, and the original conference content training data set includes: multiple groups of conference content training data, and the conference content training data includes: original conference content questions, original conference content search fragment sequences, and original conference content answers.

[0160] It should be noted that the acquisition module 1101 and the query module 1102 correspond to steps S61 to S62 in Example 3, and the examples and application scenarios implemented by the two modules and the corresponding steps are the same, but are not limited to the contents disclosed in the above-mentioned Example 3. It should be noted that the above-mentioned modules or units may be hardware components or software components stored in a memory and processed by one or more processors, and the above-mentioned modules may also be run in a computer terminal as part of a device.

[0161] Fig.12 is a structural block diagram of another data processing device according to Embodiment 6 of the present application, such as Fig.12 As shown, the device comprises:

[0162] An acquisition module 1201 is used to acquire a query processing request through a first application programming interface;

[0163] The return module 1202 is used to return a query processing response through a second application programming interface; wherein the request data carried in the query processing request includes: a query question, and the response data carried in the query processing response includes: a query reply, and the query reply is obtained by query processing the query question using a target query processing model, and the target query processing model is trained based on a target training data set, and the target training data set is obtained by expanding an original training data set, and the original training data set includes: multiple groups of training data, and the training data includes: an original question, an original search fragment sequence, and an original answer.

[0164] It should be noted that the acquisition module 1201 and the return module 1202 correspond to steps S71 to S72 in Example 4, and the examples and application scenarios implemented by the two modules and the corresponding steps are the same, but are not limited to the contents disclosed in the above-mentioned Example 4. It should be noted that the above-mentioned modules or units may be hardware components or software components stored in a memory and processed by one or more processors, and the above-mentioned modules may also be run in a computer terminal as part of a device.

[0165] Fig.13 is a structural block diagram of another data processing device according to Embodiment 6 of the present application, such as Fig.13 As shown, the device comprises:

[0166] The acquisition module 1301 is used to acquire the query processing dialogue request currently input;

[0167] A return module 1302 is used to return a query processing dialogue reply in response to the query processing dialogue request;

[0168] The request data carried in the query processing dialogue request includes: a query question, and the information carried in the query processing dialogue reply includes: a query reply, which is obtained by query processing the query question using a target query processing model, the target query processing model is trained based on a target training data set, the target training data set is obtained by expanding an original training data set, the original training data set includes: multiple sets of training data, and the training data includes: an original question, an original search fragment sequence, and an original answer;

[0169] The display module 1303 is used to display the query response in the graphical user interface.

[0170] It should be noted that the acquisition module 1301, the return module 1302 and the display module 1303 correspond to steps S81 to S83 in Example 5, and the three modules and the corresponding steps implement the same examples and application scenarios, but are not limited to the contents disclosed in the above-mentioned Example 5. It should be noted that the above-mentioned modules or units may be hardware components or software components stored in a memory and processed by one or more processors, and the above-mentioned modules may also be part of a device and run in a computer terminal.

[0171] It should be noted that the preferred implementation scheme involved in the above embodiments of the present application is the same as the scheme provided in Example 1, as well as the application scenario and implementation process, but is not limited to the scheme provided in Example 1.

[0172] Example 7

[0173] The embodiment of the present application may provide a computer terminal, which may be any computer terminal device in a computer terminal group. Optionally, in this embodiment, the computer terminal may also be replaced by a terminal device such as a mobile terminal.

[0174] Optionally, in this embodiment, the computer terminal may be located in at least one network device among a plurality of network devices of the computer network.

[0175] In this embodiment, the above-mentioned computer terminal can execute the program code of the following steps in the model training method: obtaining a target training data set, wherein the target training data set is obtained by expanding the original training data set, and the original training data set includes: multiple groups of training data, and the training data includes: original questions, original search fragment sequences and original answers; training based on the target training data set to obtain a target query processing model, wherein the target query processing model is used to perform query processing on the currently input query question to obtain a query reply.

[0176] Optionally, Fig.14 is a structural block diagram of a computer terminal according to Embodiment 7 of the present application. Fig.14As shown, the computer terminal may include: one or more (only one is shown in the figure) processors 142, a memory 144, a storage controller, and a peripheral interface, wherein the peripheral interface is connected to a radio frequency module, an audio module, and a display.

[0177] Among them, the memory can be used to store software programs and modules, such as the model training method or data processing method and device corresponding to the program instructions / modules in the embodiment of the present application, and the processor executes various functional applications and data processing by running the software programs and modules stored in the memory, that is, realizing the above-mentioned model training method or data processing method. The memory may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include a memory remotely arranged relative to the processor, and these remote memories can be connected to the computer terminal via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof.

[0178] The processor can call the information and application stored in the memory through the transmission device to perform the following steps: obtain a target training data set, wherein the target training data set is obtained by expanding the original training data set, and the original training data set includes: multiple groups of training data, and the training data includes: original questions, original search fragment sequences and original answers; obtain a target query processing model based on the target training data set training, wherein the target query processing model is used to perform query processing on the currently input query question to obtain a query reply.

[0179] Optionally, the processor may also execute the following program code: filtering the original training data set according to preset filtering conditions to obtain a filtered training data set, wherein the preset filtering conditions are set based on the degree of match between the original question and the corresponding original answer; and expanding the filtered training data set to obtain a target training data set.

[0180] Optionally, the processor may also execute the program code of the following steps: expanding the query content of the original question in the filtered training data set to obtain a target question; expanding the original search fragment sequence in the filtered training data set to obtain a target search fragment sequence; expanding the original answer based on the target question and the target search fragment sequence to obtain a target answer; obtaining a target training data set based on the target question, the target search fragment sequence and the target answer.

[0181] Optionally, the processor may also execute program code for the following steps: obtaining a first prompt text, wherein the first prompt text is used to prompt to correct the query content of the original question based on a given document, or to rewrite the query content of the original question based on the given document and correct the rewritten question; based on the first prompt text, the query content of the original question in the filtered training data set is expanded to obtain the target question.

[0182] Optionally, the processor may also execute program code for the following steps: in response to the original search fragment sequence containing multiple consecutive search fragments extracted from the same document, based on the position information of the original search fragment sequence in the same document and a preset search fragment expansion number, the number of search fragments of the original search fragment sequence in the filtered training data set is expanded to obtain a target search fragment sequence.

[0183] Optionally, the processor may also execute program code for the following steps: in response to the original search fragment sequence containing multiple search fragments extracted from multiple documents respectively, retrieving a preset number of search fragments from the target document library according to the original search fragment sequence; while retaining the original search fragment sequence, expanding the number of search fragments of the original search fragment sequence in the filtered training data set according to the comparison results between the preset number of search fragments and the original search fragment sequence and the preset search fragment expansion number, so as to obtain a target search fragment sequence.

[0184] Optionally, the processor may also execute the following program code: obtaining a first answer corresponding to the first part of the question content based on general knowledge, and obtaining a second answer corresponding to the second part of the question content based on a target search fragment sequence; and expanding the original answer by fusing the first answer and the second answer to obtain a target answer.

[0185] Optionally, the processor may also execute program code for the following steps: obtaining a second prompt text, wherein the second prompt text is used to prompt whether the target question depends on a given document and to provide a corresponding answer for a first part of the question content in the target question that does not depend on the given document; based on the second prompt text, obtaining a first answer corresponding to the first part of the question content from general knowledge.

[0186] Optionally, the processor may also execute program code for the following steps: obtaining a third prompt text, wherein the third prompt text is used to prompt retrieval of a given document and generation of an answer corresponding to the second part of the question content; based on the third prompt text, obtaining a second answer corresponding to the second part of the question content from the target search fragment sequence.

[0187] Optionally, the processor may also execute the program code of the following steps: obtaining a fourth prompt text, wherein the fourth prompt text is used to prompt the distinction of sources of answers to different parts of the target question; fusing the first answer and the second answer based on the fourth prompt text to obtain a fusion result; and expanding the original answer based on the fusion result to obtain the target answer.

[0188] The processor can call the information and application programs stored in the memory through the transmission device to execute the following steps: obtain the currently input query question; use the target query processing model to query the query question and obtain a query reply; wherein the target query processing model is trained based on a target training data set, and the target training data set is obtained by expanding the original training data set, and the original training data set includes: multiple groups of training data, and the training data includes: original questions, original search fragment sequences and original answers.

[0189] Optionally, the processor may also execute program code of the following steps: using a target query processing model to retrieve a plurality of search fragments to be used from multimedia files and / or general knowledge associated with the query question, and generating a query reply based on the plurality of search fragments to be used.

[0190] The processor can call the information and application programs stored in the memory through the transmission device to execute the following steps: obtain the currently input video conference query question; use the conference content query processing model to query and process the video conference query question to obtain a video conference query reply; wherein, the conference content query processing model is trained based on a target conference content training data set, and the target conference content training data set is obtained by expanding the original conference content training data set, and the original conference content training data set includes: multiple groups of conference content training data, and the conference content training data includes: original conference content questions, original conference content search fragment sequences and original conference content answers.

[0191] The processor can call the information and application stored in the memory through the transmission device to perform the following steps: obtain a query processing request through a first application programming interface; return a query processing response through a second application programming interface; wherein the request data carried in the query processing request includes: a query question, and the response data carried in the query processing response includes: a query reply, and the query reply is obtained by query processing the query question using a target query processing model, and the target query processing model is trained based on a target training data set, and the target training data set is obtained by expanding an original training data set, and the original training data set includes: multiple groups of training data, and the training data includes: an original question, an original search fragment sequence, and an original answer.

[0192] The processor can call the information and application stored in the memory through the transmission device to perform the following steps: obtain the currently input query processing dialogue request; return the query processing dialogue reply in response to the query processing dialogue request; wherein the request data carried in the query processing dialogue request includes: the query question, and the information carried in the query processing dialogue reply includes: the query reply, the query reply is obtained by query processing the query question using a target query processing model, the target query processing model is trained based on a target training data set, the target training data set is obtained by expanding the original training data set, the original training data set includes: multiple sets of training data, the training data includes: the original question, the original search fragment sequence and the original answer; and the query reply is displayed in a graphical user interface.

[0193] By adopting the embodiment of the present application, a target training data set is obtained by expanding the original training data set, and then a target query processing model is obtained by training based on the target training data set. The target query processing model can be used to perform query processing on the currently input query question to obtain a query reply. The accuracy and universality of the query question are judged by using the target query processing model, and then the wrong query question is corrected before answering, or a richer answer other than the original text is provided under appropriate circumstances, thereby providing more robust and richer answers, helping users to obtain more information input, and achieving the purpose of quickly obtaining accurate query replies to the currently input query questions, thereby achieving the technical effect of improving the flexibility of query processing and the accuracy of query results, and thus solving the technical problems of poor flexibility of the query processing method provided in the related art and low accuracy of the query results.

[0194] Those skilled in the art will appreciate that the structure shown in the figure is for illustration only, and the computer terminal may also be a smart phone (such as an Android phone, an iOS phone, etc.), a tablet computer, a PDA, a mobile Internet device (MID), a PAD, or other terminal devices. Fig.14 It does not limit the structure of the above electronic device. For example, the computer terminal may also include Fig.14 More or fewer components (such as network interfaces, display devices, etc.) shown in, or with Fig.14 Different configurations are shown.

[0195] A person of ordinary skill in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing the hardware related to the terminal device through a program, and the program can be stored in a computer-readable storage medium, and the storage medium may include: a flash drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0196] Example 8

[0197] The embodiment of the present application also provides a computer-readable storage medium. Optionally, in this embodiment, the storage medium can be used to store the program code executed by the model training method or data processing method provided in the first embodiment.

[0198] Optionally, in this embodiment, the above storage medium may be located in any computer terminal in a computer terminal group in a computer network, or in any mobile terminal in a mobile terminal group.

[0199] Optionally, in this embodiment, the storage medium is configured to store program code for executing the following steps: obtaining a target training data set, wherein the target training data set is obtained by expanding an original training data set, and the original training data set includes: multiple groups of training data, and the training data includes: original questions, original search fragment sequences, and original answers; obtaining a target query processing model based on training of the target training data set, wherein the target query processing model is used to perform query processing on a currently input query question to obtain a query reply.

[0200] Optionally, in this embodiment, the storage medium is configured to store program code for executing the following steps: obtaining a currently input query question; performing query processing on the query question using a target query processing model to obtain a query reply; wherein the target query processing model is trained based on a target training data set, and the target training data set is obtained by expanding an original training data set, and the original training data set includes: multiple groups of training data, and the training data includes: an original question, an original search fragment sequence, and an original answer.

[0201] Optionally, in this embodiment, the storage medium is configured to store program code for executing the following steps: obtaining the currently input video conference query question; using a conference content query processing model to query and process the video conference query question to obtain a video conference query reply; wherein the conference content query processing model is trained based on a target conference content training data set, and the target conference content training data set is obtained by expanding an original conference content training data set, and the original conference content training data set includes: multiple groups of conference content training data, and the conference content training data includes: original conference content questions, original conference content search fragment sequences, and original conference content answers.

[0202] Optionally, in this embodiment, the storage medium is configured to store program code for performing the following steps: obtaining a query processing request through a first application programming interface; returning a query processing response through a second application programming interface; wherein the request data carried in the query processing request includes: a query question, and the response data carried in the query processing response includes: a query reply, and the query reply is obtained by query processing the query question using a target query processing model, and the target query processing model is trained based on a target training data set, and the target training data set is obtained by expanding an original training data set, and the original training data set includes: multiple sets of training data, and the training data includes: an original question, an original search fragment sequence, and an original answer.

[0203] Optionally, in this embodiment, the storage medium is configured to store program code for executing the following steps: obtaining a currently input query processing dialogue request; returning a query processing dialogue reply in response to the query processing dialogue request; wherein the request data carried in the query processing dialogue request includes: a query question, and the information carried in the query processing dialogue reply includes: a query reply, and the query reply is obtained by query processing the query question using a target query processing model, the target query processing model is trained based on a target training data set, the target training data set is obtained by expanding an original training data set, and the original training data set includes: multiple sets of training data, and the training data includes: an original question, an original search fragment sequence, and an original answer; and displaying the query reply in a graphical user interface.

[0204] By adopting the embodiment of the present application, a target training data set is obtained by expanding the original training data set, and then a target query processing model is obtained by training based on the target training data set. The target query processing model can be used to perform query processing on the currently input query question to obtain a query reply. The accuracy and universality of the query question are judged by using the target query processing model, and then the wrong query question is corrected before answering, or a richer answer other than the original text is provided under appropriate circumstances, thereby providing more robust and richer answers, helping users to obtain more information input, and achieving the purpose of quickly obtaining accurate query replies to the currently input query questions, thereby achieving the technical effect of improving the flexibility of query processing and the accuracy of query results, and thus solving the technical problems of poor flexibility of the query processing method provided in the related art and low accuracy of the query results.

[0205] Example 9

[0206] According to another aspect of the embodiments of the present application, a computer program product is also provided, including a computer program, which implements the model training method or data processing method of any one of the embodiments of the present application when executed by a processor.

[0207] Optionally, the computer program included in the above-mentioned computer program product is executed by a processor to implement the following steps: obtaining a target training data set, wherein the target training data set is obtained by expanding an original training data set, and the original training data set includes: multiple groups of training data, and the training data includes: original questions, original search fragment sequences and original answers; training based on the target training data set to obtain a target query processing model, wherein the target query processing model is used to perform query processing on a currently input query question to obtain a query reply.

[0208] Optionally, the computer program included in the above-mentioned computer program product is executed by a processor to implement the following steps: obtaining a currently input query question; using a target query processing model to query process the query question to obtain a query reply; wherein the target query processing model is trained based on a target training data set, and the target training data set is obtained by expanding an original training data set, and the original training data set includes: multiple groups of training data, and the training data includes: an original question, an original search fragment sequence, and an original answer.

[0209] Optionally, the computer program included in the above-mentioned computer program product is executed by the processor to implement the following steps: obtaining the currently input video conference query question; using a conference content query processing model to query and process the video conference query question to obtain a video conference query reply; wherein the conference content query processing model is trained based on a target conference content training data set, and the target conference content training data set is obtained by expanding an original conference content training data set, and the original conference content training data set includes: multiple groups of conference content training data, and the conference content training data includes: original conference content questions, original conference content search fragment sequences and original conference content answers.

[0210] Optionally, the computer program included in the above-mentioned computer program product is executed by a processor to implement the following steps: obtaining a query processing request through a first application programming interface; returning a query processing response through a second application programming interface; wherein the request data carried in the query processing request includes: a query question, and the response data carried in the query processing response includes: a query reply, and the query reply is obtained by query processing the query question using a target query processing model, and the target query processing model is trained based on a target training data set, and the target training data set is obtained by expanding an original training data set, and the original training data set includes: multiple groups of training data, and the training data includes: an original question, an original search fragment sequence, and an original answer.

[0211] Optionally, the computer program included in the above-mentioned computer program product is executed by a processor to implement the following steps: obtaining a currently input query processing dialogue request; returning a query processing dialogue reply in response to the query processing dialogue request; wherein the request data carried in the query processing dialogue request includes: a query question, and the information carried in the query processing dialogue reply includes: a query reply, and the query reply is obtained by query processing the query question using a target query processing model, the target query processing model is trained based on a target training data set, the target training data set is obtained by expanding an original training data set, and the original training data set includes: multiple groups of training data, and the training data includes: an original question, an original search fragment sequence and an original answer; and displaying the query reply in a graphical user interface.

[0212] By adopting the embodiment of the present application, a target training data set is obtained by expanding the original training data set, and then a target query processing model is obtained by training based on the target training data set. The target query processing model can be used to perform query processing on the currently input query question to obtain a query reply. The accuracy and universality of the query question are judged by using the target query processing model, and then the wrong query question is corrected before answering, or a richer answer other than the original text is provided under appropriate circumstances, thereby providing more robust and richer answers, helping users to obtain more information input, and achieving the purpose of quickly obtaining accurate query replies to the currently input query questions, thereby achieving the technical effect of improving the flexibility of query processing and the accuracy of query results, and thus solving the technical problems of poor flexibility of the query processing method provided in the related art and low accuracy of the query results.

[0213] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0214] In the above embodiments of the present application, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0215] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0216] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0217] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0218] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, server or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, disk or optical disk, etc., various media that can store program codes.

[0219] The above is only a preferred implementation of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A model training method, characterized in that: include: Obtain a target training data set, wherein the target training data set is obtained by expanding an original training data set, the original training data set includes: multiple sets of training data, the training data includes: an original question, an original search fragment sequence and an original answer, the original search fragment sequence is used to represent information fragments related to the original question, the target training data set at least includes: a target search fragment sequence, the target search fragment sequence is obtained by expanding the original search fragment sequence by adopting a target sequence expansion method, the target sequence expansion method is determined according to a sequence source scenario corresponding to the original search fragment sequence, and the sequence source scenario includes one of the following: a single document scenario and a cross-document scenario; A target query processing model is trained based on the target training data set, wherein the target query processing model is used to perform query processing on a currently input query question to obtain a query reply.

2. The model training method according to claim 1, characterized in that: Acquiring the target training data set includes: Filtering the original training data set according to a preset screening condition to obtain a filtered training data set, wherein the preset screening condition is set based on a matching degree between the original question and the corresponding original answer; The filtered training data set is expanded to obtain the target training data set.

3. The model training method according to claim 2, characterized in that: Expanding the filtered training data set to obtain the target training data set includes: Expanding the query content of the original question in the filtered training data set to obtain a target question; Expanding the original search fragment sequence in the filtered training data set to obtain a target search fragment sequence; Expanding the original answer based on the target question and the target search fragment sequence to obtain a target answer; The target training data set is obtained according to the target question, the target search fragment sequence and the target answer.

4. The model training method according to claim 3, characterized in that: The query content of the original question in the filtered training data set is expanded to obtain the target question including: Obtaining a first prompt text, wherein the first prompt text is used to prompt to correct the query content of the original question according to a given document, or to rewrite the query content of the original question according to the given document and correct the rewritten question; The query content of the original question in the filtered training data set is expanded based on the first prompt text to obtain the target question.

5. The model training method according to claim 3, characterized in that: Expanding the original search segment sequence in the filtered training data set to obtain the target search segment sequence includes: In response to the original search fragment sequence containing multiple continuous search fragments extracted from the same document, based on the position information of the original search fragment sequence in the same document and the preset search fragment expansion number, the number of search fragments of the original search fragment sequence in the filtered training data set is expanded to obtain the target search fragment sequence.

6. The model training method according to claim 3, characterized in that: Expanding the original search segment sequence in the filtered training data set to obtain the target search segment sequence includes: In response to the original search fragment sequence comprising a plurality of search fragments respectively extracted from a plurality of documents, retrieving a preset number of search fragments from a target document library according to the original search fragment sequence; On the premise of retaining the original search fragment sequence, the number of search fragments of the original search fragment sequence in the filtered training data set is expanded according to the comparison results between the preset number of search fragments and the original search fragment sequence and the preset search fragment expansion number to obtain the target search fragment sequence.

7. The model training method according to claim 3, characterized in that: The target question includes: a first part of the question content and a second part of the question content. The original answer is expanded based on the target question and the target search fragment sequence to obtain the target answer including: Obtaining a first answer corresponding to the first part of the question content according to general knowledge, and obtaining a second answer corresponding to the second part of the question content according to the target search segment sequence; The target answer is obtained by fusing the first answer and the second answer to expand the original answer.

8. The model training method according to claim 7, characterized in that: Obtaining the first answer corresponding to the first part of the question content according to the general knowledge includes: Obtaining a second prompt text, wherein the second prompt text is used to prompt whether the target question depends on the given document and to provide a corresponding answer to the first part of the question content in the target question that does not depend on the given document; Based on the second prompt text, the first answer corresponding to the first part of the question content is obtained from the general knowledge.

9. The model training method according to claim 7, characterized in that: Acquiring the second answer corresponding to the second part of the question content according to the target search fragment sequence includes: Obtaining a third prompt text, wherein the third prompt text is used to prompt searching for a given document and generating an answer corresponding to the second part of the question content; Based on the third prompt text, the second answer corresponding to the second part of the question content is obtained from the target search fragment sequence.

10. The model training method according to claim 7, characterized in that: By integrating the first answer and the second answer, the original answer is expanded to obtain the target answer including: Obtaining a fourth prompt text, wherein the fourth prompt text is used to prompt the user to distinguish sources of answers to different parts of the target question; Based on the fourth prompt text, the first answer and the second answer are merged to obtain a fusion result; The original answer is expanded according to the fusion result to obtain the target answer.

11. A data processing method, characterized in that: include: Get the query question currently entered; Using a target query processing model to perform query processing on the query question to obtain a query response; Among them, the target query processing model is trained based on a target training data set, and the target training data set is obtained by expanding the original training data set, and the original training data set includes: multiple groups of training data, and the training data includes: original questions, original search fragment sequences and original answers, and the original search fragment sequences are used to represent information fragments related to the original questions. The target training data set at least includes: a target search fragment sequence, and the target search fragment sequence is obtained by expanding the original search fragment sequence by adopting a target sequence expansion method. The target sequence expansion method is determined according to the sequence source scenario corresponding to the original search fragment sequence, and the sequence source scenario includes one of the following: a single document scenario and a cross-document scenario.

12. The data processing method according to claim 11, characterized in that: Using the target query processing model to perform query processing on the query question to obtain the query response includes: The target query processing model is used to retrieve a plurality of search fragments to be used from multimedia files and / or general knowledge associated with the query question, and the query reply is generated based on the plurality of search fragments to be used.

13. A data processing method, characterized in that: include: Get the currently input video conference query question; Using the conference content query processing model to query the video conference query question and obtain a video conference query response; Among them, the conference content query processing model is trained based on a target conference content training data set, and the target conference content training data set is obtained by expanding the original conference content training data set, and the original conference content training data set includes: multiple groups of conference content training data, and the conference content training data includes: original conference content questions, original conference content search fragment sequences and original conference content answers, and the original conference content search fragment sequences are used to represent information fragments related to the original conference content questions. The target conference content training data set at least includes: a target conference content search fragment sequence, and the target conference content search fragment sequence is obtained by expanding the original conference content search fragment sequence by adopting a target sequence expansion method, and the target sequence expansion method is determined according to the sequence source scenario corresponding to the original conference content search fragment sequence, and the sequence source scenario includes one of the following: a single document scenario and a cross-document scenario.

14. A data processing method, characterized in that: include: Obtaining a query processing request through a first application programming interface; returning a query processing response via a second application programming interface; Among them, the request data carried in the query processing request includes: a query question, and the response data carried in the query processing response includes: a query reply, and the query reply is obtained by query processing the query question using a target query processing model, and the target query processing model is trained based on a target training data set, and the target training data set is obtained by expanding an original training data set, and the original training data set includes: multiple groups of training data, and the training data includes: an original question, an original search fragment sequence and an original answer, and the original search fragment sequence is used to represent information fragments related to the original question, and the target training data set at least includes: a target search fragment sequence, and the target search fragment sequence is obtained by expanding the original search fragment sequence using a target sequence expansion method, and the target sequence expansion method is determined according to a sequence source scenario corresponding to the original search fragment sequence, and the sequence source scenario includes one of the following: a single document scenario and a cross-document scenario.

15. A data processing method, characterized in that: include: Get the query processing dialog request currently entered; In response to the query processing dialog request, returning a query processing dialog reply; The request data carried in the query processing dialogue request includes: a query question; the information carried in the query processing dialogue reply includes: a query reply, the query reply is obtained by query processing the query question using a target query processing model; the target query processing model is trained based on a target training data set; the target training data set is obtained by expanding an original training data set; the original training data set includes: multiple sets of training data; the training data includes: an original question, an original search fragment sequence, and an original answer; the original search fragment sequence is used to represent information fragments related to the original question; the target training data set includes at least: a target search fragment sequence; the target search fragment sequence is obtained by expanding the original search fragment sequence using a target sequence expansion method; the target sequence expansion method is determined according to a sequence source scenario corresponding to the original search fragment sequence; the sequence source scenario includes one of the following: a single document scenario, a cross-document scenario; The query response is presented within a graphical user interface.

16. An electronic device, characterized in that: include: A memory storing an executable program; A processor, used to run the program, wherein the program, when running, executes the model training method described in any one of claims 1 to 10 or the data processing method described in any one of claims 11 to 15.

17. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored executable program, wherein, when the executable program is running, the device where the computer-readable storage medium is located is controlled to execute the model training method described in any one of claims 1 to 10 or the data processing method described in any one of claims 11 to 15.

18. A computer program product, characterized in that It includes a computer program which, when executed by a processor, implements the model training method described in any one of claims 1 to 10 or the data processing method described in any one of claims 11 to 15.

Images