Language processing model training method and device and question processing method and device

Abstract

The embodiment of the invention provides a language processing model training method and device and a question processing method and device.The language processing model training method comprises the steps that a target text is determined, and a text question corresponding to the target text is generated; generating a prompt text from the target text, the text question corresponding to the target text and the text key point, and inputting the prompt text into an initial language processing model to obtain the text key point corresponding to the text question; according to the target text, the text question corresponding to the target text and the text key point corresponding to the text question, constructing text training data; and training the initial language processing model according to the text training data to obtain a target language processing model. Under the guidance of the text problem, the text key points strongly related to the text problem are generated through the large language model, and model fine tuning iteration is performed, so that the efficiency and accuracy of subsequent search of the large language model according to the input problem are enhanced.

Description

Technical Field

[0001] The embodiments of this specification relate to the field of computer technology, and in particular to a language processing model training method and device, a problem processing method and device. One or more embodiments of this specification also relate to a medical problem processing method and device, a computing device, a computer-readable storage medium, and a computer program product. Background Art

[0002] In the era of information explosion, the rapid growth of massive data has brought great challenges to people in information screening and processing. Especially in some professional scientific research fields (such as medical research), with the surge in the number of scientific research documents, researchers are faced with the problem of how to accurately and efficiently extract key information from numerous scientific research documents.

[0003] Traditional information retrieval methods often find it difficult to cope with this challenge because, although the overly long text of scientific research documents contains rich background and details, it also leads to a large amount of information redundancy, which reduces reading efficiency. Then this traditional information retrieval method cannot quickly locate the truly important document content, which in turn affects the retrieval accuracy and efficiency of extracting key information from numerous scientific research documents.

[0004] Therefore, how to quickly locate key information from a large amount of text data and improve the retrieval accuracy and efficiency of key information has become an urgent problem to be solved. Summary of the invention

[0005] In view of this, the embodiments of this specification provide a language processing model training method and device, and a problem processing method and device. One or more embodiments of this specification also relate to a medical problem processing method and device, a computing device, a computer-readable storage medium, and a computer program product to solve the technical defects in the prior art that key information cannot be quickly located from a large amount of text data, reducing the accuracy and efficiency of key information retrieval.

[0006] According to a first aspect of an embodiment of this specification, a language processing model training method is provided, comprising:

[0007] Determine a target text, and generate text questions corresponding to the target text;

[0008] The target text, the text question corresponding to the target text, and the text key points generate prompt text, which are input into the initial language processing model to obtain the text key points corresponding to the text question;

[0009] Constructing text training data according to the target text, the text question corresponding to the target text, and the text key points corresponding to the text question;

[0010] The initial language processing model is trained according to the text training data to obtain a target language processing model.

[0011] According to a second aspect of the embodiments of this specification, a language processing model training device is provided, including:

[0012] A determination module is configured to determine a target text and generate a text question corresponding to the target text;

[0013] A key point acquisition module is configured to input the target text, the text question corresponding to the target text, and the prompt text generated by the text key points into an initial language processing model to obtain the text key points corresponding to the text question;

[0014] A training data construction module is configured to construct text training data according to the target text, the text question corresponding to the target text, and the text key points corresponding to the text question;

[0015] The training module is configured to train the initial language processing model according to the text training data to obtain a target language processing model.

[0016] According to a third aspect of the embodiments of this specification, a problem handling method is provided, including:

[0017] Receiving a target question, inputting the target question into a target language processing model, and obtaining text key points corresponding to the target question, wherein the target language processing model is trained according to the above-mentioned language processing model training method;

[0018] According to the main points of the text, determining the target document corresponding to the main points of the text;

[0019] The target question, the text main points, and the target document are input into the target language processing model to obtain a target answer corresponding to the target question.

[0020] According to a fourth aspect of the embodiments of this specification, there is provided a problem handling device, including:

[0021] A key point acquisition module is configured to receive a target question, input the target question into a target language processing model, and obtain text key points corresponding to the target question, wherein the target language processing model is obtained by training according to the above-mentioned language processing model training method;

[0022] A document determination module is configured to determine a target document corresponding to the text key point according to the text key point;

[0023] The answer generation module is configured to input the target question, the text key points, and the target document into the target language processing model to obtain a target answer corresponding to the target question.

[0024] According to a fifth aspect of the embodiments of this specification, a method for handling medical problems is provided, comprising:

[0025] Receiving a target medical question, inputting the target medical question into a target language processing model, and obtaining key points of a medical text corresponding to the target medical question, wherein the target language processing model is trained according to the above-mentioned language processing model training method;

[0026] According to the main points of the medical text, determining a target medical document corresponding to the main points of the medical text;

[0027] The target medical question, the main points of the medical text, and the target medical document are input into the target language processing model to obtain a target medical answer corresponding to the target medical question.

[0028] According to a sixth aspect of the embodiments of this specification, there is provided a medical problem processing device, comprising:

[0029] A medical key point acquisition module is configured to receive a target medical question, input the target medical question into a target language processing model, and obtain the medical text key points corresponding to the target medical question, wherein the target language processing model is obtained by training according to the above-mentioned language processing model training method;

[0030] A medical document determination module is configured to determine a target medical document corresponding to the medical text key points according to the medical text key points;

[0031] The medical answer generation module is configured to input the target medical question, the main points of the medical text, and the target medical document into the target language processing model to obtain a target medical answer corresponding to the target medical question.

[0032] According to a seventh aspect of an embodiment of this specification, there is provided a computing device, including:

[0033] Memory and processor;

[0034] Among them, the memory is used to store computer programs / instructions, and the processor is used to execute computer programs / instructions. When the computer program / instructions are executed by the processor, the steps of the above-mentioned language processing model training method, problem processing method, or medical problem processing method are implemented.

[0035] According to the eighth aspect of the embodiments of this specification, a computer-readable storage medium is provided, which stores a computer program / instruction, and when the computer program / instruction is executed by a processor, the steps of the above-mentioned language processing model training method, problem processing method, or medical problem processing method are implemented.

[0036] According to the ninth aspect of the embodiments of this specification, a computer program product is provided, including a computer program / instruction, which, when executed by a processor, implements the steps of the above-mentioned language processing model training method, problem processing method, or medical problem processing method.

[0037] An embodiment of the present specification implements a language processing model training method. The method generates a prompt text according to the text question corresponding to the target text and in combination with the text key points, guides the initial language processing model to generate the text key points corresponding to the text question based on the target text, and enables the initial language processing model to better understand the true intention of the inquirer according to the text question, thereby being able to dig out the potential information hidden under the surface text of the target text. At the same time, the initial language processing model pays more attention to semantic understanding, and through semantic analysis, identifies the relationship between words in the target text, the meaning of sentences and the overall intention of the target text, thereby extracting higher-level information, and further makes the extracted text key points strongly correlated with the text question corresponding to the target text.

[0038] Then, according to the target text, the text question corresponding to the target text, and the text key points strongly related to the text question, text training data is constructed, and the initial language processing model is trained to obtain the target language processing model; in the specific scenario in which the target language processing model is subsequently applied, when a certain target question is input, the text key points corresponding to the target question can be quickly and accurately obtained according to the target language processing model, so that according to the correspondence between the text key points and the target text, the target text corresponding to the target question can be quickly and accurately located from a large amount of text data, and the question can be answered, thereby improving the retrieval accuracy and efficiency of key information. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is a scenario diagram of a language processing model training method provided by an embodiment of this specification;

[0040] Figure 2 is a flow chart of a language processing model training method provided by one embodiment of this specification;

[0041] Figure 3 is a flowchart of determining a target text in a language processing model training method provided by an embodiment of this specification;

[0042] Figure 4 It is a flow chart of the overall processing process of a language processing model training method provided by one embodiment of this specification;

[0043] Figure 5 It is a processing flow chart of a language processing model training method in the medical field provided by an embodiment of this specification;

[0044] Figure 6 is a flow chart of a problem handling method provided by an embodiment of this specification;

[0045] Figure 7 is a flow chart of a method for handling a medical problem provided by one embodiment of this specification;

[0046] Figure 8 It is a structural schematic diagram of a language processing model training device provided by an embodiment of this specification;

[0047] Fig. 9 It is a structural schematic diagram of a problem handling device provided by an embodiment of this specification;

[0048] Fig.10 It is a structural schematic diagram of a medical problem treatment device provided by one embodiment of this specification;

[0049] Fig.11 It is a structural block diagram of a computing device provided by an embodiment of this specification. DETAILED DESCRIPTION

[0050] Many specific details are described in the following description to facilitate a full understanding of this specification. However, this specification can be implemented in many other ways than those described herein, and those skilled in the art can make similar generalizations without violating the connotation of this specification, so this specification is not limited to the specific implementation disclosed below.

[0051] The terms used in one or more embodiments of this specification are only for the purpose of describing specific embodiments, and are not intended to limit one or more embodiments of this specification. The singular forms of "a", "said" and "the" used in one or more embodiments of this specification and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used in one or more embodiments of this specification refers to and includes any or all possible combinations of one or more associated listed items.

[0052] It should be understood that although the terms first, second, etc. may be used to describe various information in one or more embodiments of this specification, this information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of one or more embodiments of this specification, the first may also be referred to as the second, and similarly, the second may also be referred to as the first. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".

[0053] In addition, 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 one or more embodiments of this specification 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 the relevant countries and regions, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0054] In one or more embodiments of this specification, a large model refers to a deep learning model with large-scale model parameters, which usually contains hundreds of millions, tens of billions, hundreds of billions, trillions, or even more than 10 trillion model parameters. A large model can also be called a foundation model / foundation model (Foundation Model 1). 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 a large-scale language model (LLM, Large Language Model 1), a multi-modal pre-training model (multi-modal pre-training model 1), etc.

[0055] When the big model is used in practice, only a small number of samples are needed to fine-tune the pre-trained model and it can be applied to different tasks. The big model can be widely used in natural language processing (NLP), computer vision and other fields. Specifically, it can be applied to computer vision tasks such as visual question answering (VQA), image description (IC), image generation, as well as natural language processing tasks such as text-based sentiment classification, text summary generation, and machine translation. The main application scenarios of the big model include digital assistants, intelligent robots, search, online education, office software, e-commerce, intelligent design, etc.

[0056] First, the terms involved in one or more embodiments of this specification are explained.

[0057] Text highlights: Extract representative and important content from a large amount of information in the text. These contents can effectively summarize the main idea, key information and internal logic of the text. Text highlights not only help to quickly understand the text, but also support deeper analysis and decision-making. Text highlights emphasize importance, relevance and operability, ensuring that the extracted information can provide practical value to users, avoiding redundancy and noise, and making information processing more efficient and accurate.

[0058] Prompt: In natural language processing, a prompt is an input method that guides the model to generate specific output through specific instructions or questions; it can be an instruction, question, or part of a sentence.

[0059] Large language model: A deep learning model trained with large amounts of text data that can generate natural language text or understand the meaning of language text. Large language models can handle a variety of natural language tasks, such as text classification, question answering, and conversation.

[0060] To solve the above technical problems, in this specification, a language processing model training method and device, a problem processing method and device are provided. One or more embodiments of this specification also involve a medical problem processing method and device, a computing device, a computer-readable storage medium, and a computer program product, which are described in detail one by one in the following embodiments.

[0061] See also Figure 1 , Figure 1 A scenario diagram of a language processing model training method provided by an embodiment of the present specification is shown.

[0062] Specifically, the language processing model training method is implemented using an end-side device 102 and a server 104. The end-side device 102 is used to send a model training request for an initial language processing model applied to a target field to the server 104. For example, the model training request is "training a language processing model for the field of medical research". In actual applications, users can input model training requests in the end-side device 102 in the form of text or voice. If voice is used, the end-side device 102 will also include corresponding voice processing parts, such as voice analysis, voice-to-text, voice synthesis and other modules, which are used to convert the model training request input by the user through voice into text. This manual does not impose any restrictions on this.

[0063] Taking the model training request of "training a language processing model for the field of medical research" as an example, the server 104 responds to the model training request sent by the terminal device 102, and obtains the target text in the field of medical research that has been semantically segmented and structured from the semantic database through the language processing model training method, and generates text questions corresponding to the target text; generates prompt text for the target text, the text questions corresponding to the target text, and the text key points, and inputs the prompt text into the initial language processing model to obtain the text key points corresponding to the text questions; constructs text training data according to the target text, the text questions corresponding to the target text, and the text key points corresponding to the text questions; trains the initial language processing model according to the text training data to obtain the target language processing model.

[0064] After the training of the target language processing model is completed, the server 104 can send the target language processing model to the terminal device 102 for deployment, or return the calling interface of the target language processing model to the terminal device 102 to enable the terminal device 102 to apply the target language processing model. For example, when the target language processing model is applied to a question-and-answer scenario in the field of medical research, when the terminal device 102 responds to a medical question input by a user, the target language processing model can be used to obtain the key points of the medical text corresponding to the medical question, thereby quickly and accurately locating the medical text corresponding to the medical question from a large amount of medical text data based on the correspondence between the key points of the medical text and the medical text, so as to generate an answer based on the key points of the medical text and the medical text corresponding to the medical question.

[0065] The end-side device 102 may include a browser, an APP (Application), or a web application such as an H5 (Hyper Text Markup Language 5, version 5 of Hypertext Markup Language) application, or a light application (also known as a mini-program, a lightweight application) or a cloud application, etc. The end-side device may be based on the software development kit (SDK, Software Development Kit) of the corresponding service provided by the server, such as based on the real-time communication (RTC, Real Time Communication) SDK development and acquisition, etc. The end-side device may be deployed in an electronic device and needs to rely on the device to run or some APPs in the device to run, etc. The electronic device may have a display screen and support information browsing, etc., such as a personal mobile terminal such as a mobile phone, a tablet computer, a personal computer, etc. Various other types of applications may also be configured in the electronic device, such as human-computer dialogue applications, model training applications, data processing applications, web browser applications, shopping applications, search applications, instant messaging tools, email clients, social platform software, etc.

[0066] Server 104 can be understood as a server that provides various services, including physical servers and cloud servers, such as a server that provides communication services to multiple clients, a server for background training that provides support for models used on clients, and a server that processes data sent by clients. It should be noted that server 104 can be implemented as a distributed server cluster consisting of multiple servers, or as a single server. Server 104 can also be a server for a distributed system, or a server combined with a blockchain. Server 104 can also be a cloud server for basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content distribution networks (CDN, Content De l ivery Network), and big data and artificial intelligence platforms, or an intelligent cloud computing server or intelligent cloud host with artificial intelligence technology.

[0067] It is worth noting that the language processing model training method provided in the embodiments of this specification can be executed by the server 104. In other embodiments of this specification, when the computing resources of the terminal device 102 are relatively large, the terminal device 102 and the server 104 can also have similar functions, thereby executing the language processing model training method provided in the embodiments of this specification; in other embodiments, the language processing model training method provided in the embodiments of this specification can also be executed jointly by the terminal device 102 and the server 104.

[0068] An embodiment of the present specification provides a language processing model training method. After determining a target text and a text question corresponding to the target text, the method generates a prompt text according to the text question corresponding to the target text and in combination with text key points, guides the initial language processing model to generate text key points corresponding to the text question based on the target text, so that the initial language processing model can better understand the true intention of the inquirer according to the text question, thereby being able to dig out the potential information hidden under the surface text of the target text, and at the same time, the initial language processing model pays more attention to semantic understanding, and through semantic analysis, identifies the relationship between words in the target text, the meaning of sentences and the overall intention of the target text, thereby extracting higher-level information, and further making the extracted text key points strongly correlated with the text question corresponding to the target text.

[0069] Then, according to the target text, the text question corresponding to the target text, and the text key points strongly related to the text question, text training data is constructed, and the initial language processing model is trained to obtain the target language processing model; in the specific scenario in which the target language processing model is subsequently applied, when a certain target question is input, the text key points corresponding to the target question can be quickly and accurately obtained according to the target language processing model, so that according to the correspondence between the text key points and the target text, the target text corresponding to the target question can be quickly and accurately located from a large amount of text data, and the question can be answered, thereby improving the retrieval accuracy and efficiency of key information.

[0070] See also Figure 2 , Figure 2 A flow chart of a language processing model training method provided by an embodiment of the present specification is shown, which specifically includes the following steps.

[0071] Step 202: Determine the target text and generate text questions corresponding to the target text.

[0072] The target text may be understood as a text in any field and any format; for example, a text in the field of medical research, in PDF format or TXT format; that is, the target text may be a target text in the field of medical research.

[0073] The text questions corresponding to the target text can be understood as text questions related to the target text; for example, when the target text is a text in the field of medical research, the text questions corresponding to the target text can be medical research questions, such as "How to eat for diabetes" and the like.

[0074] In practical applications, in order to improve the efficiency of generating text key points based on the target text, the target text can be determined by a shorter text segment formed by semantic segmentation of the initial text. The specific implementation method is as follows:

[0075] The determining of the target text comprises:

[0076] An initial text is obtained, and semantic segmentation is performed on the initial text to determine the target text.

[0077] The initial text can be understood as a large document in any field and any format, for example, a medical document in the field of medical research.

[0078] Semantic segmentation of the initial text can be achieved through any segmentation method, such as sentence-level segmentation (such as using periods, greetings, and exclamation marks as sentence boundary markers for segmentation), paragraph-level segmentation (such as using blank lines, carriage returns, etc. as paragraph boundary markers for segmentation), semantic unit segmentation (such as segmentation after semantic analysis using a large language model), etc.

[0079] Specifically, taking the initial text as a medical document in the field of medical research as an example, the initial text is obtained, and the initial text is semantically segmented to determine the target text. This can be understood as obtaining multiple medical documents in the field of scientific research, and semantically segmenting each medical document, and using the segmented text segments obtained after segmentation as the target text.

[0080] In the language processing model training method provided in one or more embodiments of the present specification, after obtaining the initial text, the large-scale initial text is semantically segmented by any semantic segmentation method, and the shorter segmented text segments are used as the target text, which can greatly improve the efficiency of subsequently generating text key points based on the target text.

[0081] In order to improve the efficiency of text segmentation, avoid information redundancy in the target text after segmentation, and ensure the semantic coherence and integrity of the text, a large language model can be used to perform semantic analysis before segmentation. The specific implementation method is as follows:

[0082] The semantic segmentation of the initial text to determine the target text includes:

[0083] Inputting the initial text and the semantic segmentation prompt text into the initial language processing model to obtain a plurality of semantic text segments;

[0084] Text processing is performed on the multiple semantic text segments to determine the target text.

[0085] Among them, the initial language processing model can be understood as a large language model before training is completed.

[0086] Still using the above example, the initial text and semantic segmentation prompt text are input into the initial language processing model to obtain multiple semantic text fragments. It can be understood that each medical document in the medical document in the field of scientific research is input into the initial language processing model in combination with the semantic segmentation prompt text, and the initial language processing model is used to obtain multiple semantic text fragments corresponding to each medical document; then, text processing is performed on the multiple semantic text fragments corresponding to each medical document to determine the target text.

[0087] In the language processing model training method provided in one or more embodiments of the present specification, after obtaining the initial text, the initial language processing model is used to perform semantic segmentation on the initial text to obtain multiple semantic text fragments. Through the strong semantic understanding and generation capabilities of the large language model, on the basis of improving the accuracy and efficiency of text processing, the key information in the initial text can be better extracted and utilized, and the semantic integrity and coherence of each semantic text fragment obtained by segmentation can be ensured; thereby ensuring that text processing is performed on multiple semantic text fragments, and the determined target text does not have information redundancy, and has semantic integrity and coherence.

[0088] Then, when the target text is used for the initial language processing model training, since the target text only contains necessary information, the amount of data processed by the model is reduced, which can significantly improve the training efficiency of the model; and due to the semantic integrity of the target text, the model can better understand the meaning of the target text, thereby improving its performance; at the same time, the non-redundant target text reduces the model's dependence on specific expressions and enhances its generalization ability.

[0089] In order to further improve the overall processing efficiency and quality of semantic segmentation of the initial text using the initial language processing model, when the initial text is segmented using the initial language processing model, the initial text is first segmented according to the preset text segmentation rules, and then the initial language processing model is used to quickly and accurately segment the initial text fragments after the initial segmentation; the specific implementation method is as follows:

[0090] The initial text and the semantic segmentation prompt text are input into the initial language processing model to obtain a plurality of semantic text segments, including:

[0091] Segmenting the initial text according to a preset text segmentation rule to obtain a plurality of initial text segments;

[0092] The multiple initial text segments and the semantic segmentation prompt text are input into the initial language processing model to obtain the multiple semantic text segments.

[0093] Among them, the preset text segmentation rules can be set according to actual applications; for example, text segmentation rules set according to syntax, semantics, specific symbols, etc., such as text segmentation rules set based on line break symbols, text segmentation rules set based on headlines, etc.

[0094] Taking the preset text segmentation rule as a text segmentation rule based on line break symbol setting as an example, the initial text is segmented according to the preset text segmentation rule to obtain multiple initial text fragments. It can be understood as the text segmentation rule based on line break symbol setting, segmenting at each line break symbol position of the initial text to obtain multiple initial text fragments after segmentation.

[0095] Then, each initial text segment and semantic segmentation prompt text obtained after the preliminary segmentation are input into the initial language processing model to obtain multiple semantic text segments corresponding to each initial text segment.

[0096] In the language processing model training method provided in one or more embodiments of the present specification, before using the initial language processing model to perform semantic segmentation on the initial text, the initial text is first segmented according to preset text segmentation rules to divide the long text into multiple shorter initial text segments, thereby reducing the amount of text processing required for subsequent semantic segmentation of the initial text segments using the initial language processing model to obtain multiple semantic text segments, thereby improving the model processing efficiency. The preliminary segmentation of the initial text can also help identify and remove redundant information or irrelevant content in the initial text, and can also ensure the uniformity of the text formats of the multiple initial text segments obtained after segmenting the initial text, thereby further improving the text quality of obtaining multiple semantic text segments by semantically segmenting the initial text segments using the initial language processing model.

[0097] In order to ensure the semantic accuracy of the multiple semantic text segments obtained by semantically segmenting the initial text or the initial text segment using the initial language processing model, so as to ensure the text quality of the target text; when performing text processing on the multiple semantic text segments and determining the target text, the multiple semantic text segments can be verified using the initial language processing model to obtain high-quality semantic text segments as the target text; the specific implementation method is as follows:

[0098] The performing text processing on the multiple semantic text segments to determine the target text includes:

[0099] Inputting the plurality of semantic text segments and text verification prompt text into the initial language processing model to obtain a verification result of each semantic text segment in the plurality of semantic text segments;

[0100] According to the verification results of the semantic text segments, the multiple semantic text segments are filtered to obtain filtered target semantic text segments;

[0101] The target semantic text segment is determined as the target text.

[0102] The verification result of each semantic text segment can be understood as a semantically complete or semantically incomplete verification label, a semantically overlapping or semantically non-overlapping verification label, and the like.

[0103] Specifically, each semantic text fragment among the multiple semantic text fragments is input into an initial language processing model in combination with a text verification prompt text, and the verification result of each semantic text fragment is obtained using the initial language processing model; then, based on the verification result of each semantic text fragment, the multiple semantic text fragments are filtered, such as filtering out semantic text fragments whose verification results are semantically incomplete verification labels, and determining the remaining semantic text fragments among the multiple semantic text fragments as filtered target semantic text fragments; finally, the filtered one or more target semantic text fragments are used as target texts.

[0104] In the language processing model training method provided by one or more embodiments of the present specification, after obtaining multiple semantic text fragments using an initial language processing model, the multiple semantic text fragments and text verification prompt text are input into the initial language processing model, and the multiple semantic text fragments are semantically verified using the initial language processing model. According to the semantic verification results of the multiple semantic text fragments, the semantic text fragments with semantic overlap and / or semantic loss are filtered, and the filtered target semantic text fragments are determined as the target text for subsequent processing; thereby ensuring that the target text obtained by semantic segmentation is semantically meaningful and can independently express a complete concept or information, without semantic overlap and semantic loss, thereby ensuring the text quality of the target text.

[0105] In addition, in order to facilitate subsequent query and analysis, the target text will be structured and then stored in the semantic database. The specific implementation method is as follows:

[0106] Determining the target semantic text segment as the target text includes:

[0107] Performing structural processing on the target semantic text segment, and determining the target semantic text segment after structural processing as the target text;

[0108] After determining the target semantic text segment after the structured processing as the target text, the method further includes:

[0109] The target text is stored in a semantic database.

[0110] Specifically, after obtaining the filtered target semantic text fragment, the target semantic text fragment is structured and output as structured data, which can be in the form of key-value pairs, tables, database records, etc., to facilitate subsequent database storage; then, after the structured target semantic text fragment is determined as the target text, the target text can be stored in the semantic database to facilitate query and analysis of the target text from the semantic database.

[0111] In one or more embodiments of the present specification, the target text may be a semantic text segment obtained after semantic segmentation of the initial text, or a semantic text segment obtained by applying preset text segmentation rules and semantic segmentation to the initial text, or a semantic text segment obtained after semantic segmentation and verification of the initial text, or a semantic text segment obtained by applying preset text segmentation rules and semantic segmentation and verification to the initial text; the target text may be understood as the target text obtained from the semantic database and determined by the above-mentioned text processing method.

[0112] See also Figure 3 , Figure 3 It is a flowchart for determining a target text in a language processing model training method provided by an embodiment of this specification.

[0113] Step 302: Acquire multiple initial texts.

[0114] Specifically, obtaining multiple initial texts can be understood as obtaining multiple initial texts in a target scientific research field (such as a medical scientific research field).

[0115] Step 304: Text rule division.

[0116] Specifically, text rule segmentation can be understood as segmenting each of the multiple initial texts according to a preset text segmentation rule to obtain multiple initial text segments corresponding to each initial text.

[0117] Step 306: semantic segmentation.

[0118] Specifically, semantic segmentation can be understood as inputting each of the multiple initial text segments into the initial language processing model (i.e., the large language model) in combination with the semantic segmentation prompt (i.e., the semantic segmentation prompt text), and using the initial language processing model to perform semantic segmentation on each initial text segment to obtain multiple semantic text segments corresponding to each initial text segment.

[0119] Step 308: Output multiple semantic text segments.

[0120] Specifically, outputting a plurality of semantic text segments can be understood as the initial language processing model performing semantic segmentation on each initial text segment and then outputting a plurality of semantic text segments corresponding to each initial text segment.

[0121] Step 310: Filtering the text after semantic segmentation.

[0122] Specifically, text filtering after semantic segmentation can be understood as inputting each semantic text fragment among multiple semantic text fragments into the initial language processing model in combination with the verification prompt (i.e., the text verification prompt text), and using the initial language processing model to verify each semantic text fragment, check whether the semantics of each semantic text fragment overlap, whether the semantics are complete, etc., and output the verification result of each semantic text fragment; according to the verification result of each semantic text fragment, the semantic text fragments with overlapping or incomplete semantics are filtered out.

[0123] Step 312: Obtain the filtered target semantic text segment.

[0124] Specifically, obtaining the filtered target semantic text segment can be understood as retaining the filtered semantic text segment with no semantic overlap and complete semantics as the target semantic text segment.

[0125] Step 314: Data storage.

[0126] Specifically, data storage can be understood as storing the target semantic text segment in a semantic database.

[0127] In practical applications, in order to facilitate subsequent storage, query and analysis, the initial language processing model can process the output of multiple semantic text fragments after semantic segmentation, or the target semantic text fragments after semantic verification, and then output them as structured data; of course, in order to save computing resources, structured processing and output can also be performed when the target semantic text fragments are output at the end, and the embodiments of this specification do not impose any limitations on this.

[0128] In the language processing model training method provided in one or more embodiments of the present specification, the initial text is preliminarily segmented using a preset text segmentation rule, and then semantically segmented using an initial language processing model. The segmented semantic text segments are then semantically verified using a verification prompt. This allows unstructured text data to be converted into structured data while ensuring that each structured target semantic text segment does not have semantic overlap or semantic loss, which not only facilitates the storage of target semantic text segments, but also helps improve the training efficiency and quality of subsequent model training using the target semantic text segments.

[0129] In addition, in order to discover the common needs and preferences of users in the target field and design more reasonable text questions for the target text, we can analyze the user's historical questions and consider the user's question intentions to generate text questions that are more in line with the target text. The specific implementation method is as follows:

[0130] The generating of the text question corresponding to the target text includes:

[0131] The collected historical questions are parsed, and prompt text is generated based on the question construction result. The prompt text is generated based on the target text and the question, and a text question corresponding to the target text is generated.

[0132] Among them, historical issues can be understood as issues belonging to the same field as the target text; for example, when the target text belongs to the field of medical research, historical issues can be understood as historical issues in the field of medical research.

[0133] Specifically, the collected historical questions are parsed, and a question generation prompt text is constructed based on the parsing results. This can be understood as collecting multiple historical questions, parsing the historical texts, and generating a question generation prompt text that can better meet user expectations based on the parsing results; then generating a prompt text based on the target text and the question, and generating a text question corresponding to the target text.

[0134] In the language processing model training method provided in one or more embodiments of the present specification, by analyzing the collected historical questions, the common needs and preferences of users are determined based on the analysis results, thereby generating a question generation prompt text that can better meet the user's expectations; and a reasonably constructed question generation prompt text can provide a reference and guidance for text question generation for the target text. Through the question generation prompt text, a text question that better matches the target text can be generated, and the required information or solutions can be more easily obtained later, thereby enhancing the user experience.

[0135] The analysis of historical issues can be achieved through a variety of analysis methods, such as trained intent analysis models, rule-based intent analysis reasoning algorithms, etc. The specific implementation methods are as follows:

[0136] The collected historical questions are analyzed, and the question generation prompt text is constructed according to the analysis results, including:

[0137] Collect historical problems, perform multi-dimensional data analysis on the historical problems, and construct question generation prompt texts based on the multi-dimensional data analysis results.

[0138] Among them, multi-dimensional data analysis can be set according to actual applications, such as keyword analysis, topic analysis, semantic analysis, etc.

[0139] Taking multi-dimensional data analysis including keyword analysis, topic analysis, and semantic analysis as an example, historical issues are collected, multi-dimensional data analysis is performed on the historical issues, and based on the results of the multi-dimensional data analysis, question generation prompt text is constructed. This can be understood as collecting historical issues and pre-processing the historical issue data, including cleaning, de-duplication, and stop word removal to ensure data quality; using natural language processing technology to extract keywords from historical issues, perform part-of-speech tagging on the keywords, and distinguish different parts of speech such as nouns, verbs, and adjectives; using topic models to perform topic analysis on historical issues, identify common topic categories, and Based on the analysis results of the above three dimensions, similar questions are classified into the same category; semantic analysis technology is used to semantically represent the questions and the semantic features of the questions are extracted; according to the analysis results of the above three dimensions, a prompt text template is designed, which can contain information such as keywords and topic categories to guide the subsequent model to generate related questions; the extracted keywords are then integrated into the prompt text template to ensure that the prompt text can cover important keywords, and the identified topic categories are integrated into the prompt text template to ensure that the prompt text can guide the model to generate questions related to the topic. At this point, the designed prompt text template can be used to generate prompt text for specific questions.

[0140] In the language processing model training method provided in one or more embodiments of the present specification, by analyzing historical questions, an effective prompt text can be constructed to help the subsequent large language model generate relevant and accurate questions for the target text; and through the multi-dimensional data analysis results, it can be ensured that the generated text questions both meet the user's intentions and improve the performance and practicality of the model.

[0141] In addition, in order to improve the relevance and quality of the generated prompt text for question generation based on the analysis results, in the embodiments of this specification, the historical questions can also be analyzed based on the analysis of the real positive samples and the question analysis strategy obtained by the analysis, so as to generate high-quality prompt text for question generation based on the analysis results. The specific implementation method is as follows:

[0142] The collected historical questions are analyzed, and the question generation prompt text is constructed according to the analysis results, including:

[0143] Collect historical questions, and obtain positive sample questions corresponding to the historical questions and recall texts corresponding to the positive sample questions;

[0144] Performing sample analysis on the positive sample questions and the recalled texts corresponding to the positive sample questions to obtain a question parsing strategy;

[0145] According to the problem analysis strategy, the historical problem is analyzed, and according to the analysis result, the question is constructed to generate prompt text.

[0146] Among them, the positive sample problem can be understood as a real problem in the target field that is constructed or collected in the same target field as the historical text; for example, the positive sample problem is: please find the latest clinical trial research report on disease a.

[0147] The recalled text corresponding to the positive sample question can be understood as the real document corresponding to the positive sample question and needing to be recalled; that is, the real document related to the positive sample question retrieved according to the positive sample question.

[0148] Sample analysis of the positive sample questions and the recall texts corresponding to the positive sample questions can be understood as analyzing the advantages and disadvantages of the positive sample questions and finding out the problems therein, such as some information is not clear enough, there is too much irrelevant content in the document, and some key details are missing.

[0149] The question parsing strategy can be understood as a question parsing strategy determined based on sample analysis results, such as a keyword classification sampling strategy, which is to classify documents according to the keywords in the user's question and sample each category of keywords separately.

[0150] Taking the problem parsing strategy as a keyword classification sampling strategy as an example, according to the problem parsing strategy, the historical problem is parsed, and according to the parsing result, a question generation prompt text is generated; it can be understood that according to the keyword classification sampling strategy, the keywords in the historical questions are sampled, and by analyzing the keywords in the historical questions, a Query generation Prompt (i.e., the question generation prompt text) is generated according to the analyzed keyword classification.

[0151] For example, the historical questions are: "The latest treatment method for disease a", "Clinical trial results for disease a", "Treatment progress for disease b"; according to the keyword classification sampling strategy, keywords are extracted from the historical questions: that is, keywords are extracted from the first historical question: "latest", "disease a", "treatment method", and keywords are extracted from the second historical question: "disease a", "clinical trial", "results", and keywords are extracted from the third historical question: "disease b", "treatment progress"; the extracted keywords are classified according to the dimensions of subject, intent, and field; for example: subject: disease a, disease b, etc., intent: latest research, treatment method, clinical trial, etc., field: medicine, technology, education, etc. Finally, based on the classified keywords, a query generation prompt is generated.

[0152] In actual applications, the Query Generation Prompt includes the following elements: analysis category (such as subject category, intent category, domain category, etc.), Query Generation Prompt (a prompt used to guide the generation of new questions, which can be customized according to the analysis category to ensure that the generated questions match the user's query intent).

[0153] In the language processing model training method provided in one or more embodiments of the present specification, by performing sample analysis on positive sample questions and recall texts corresponding to the positive sample questions, the recall strategy of the recall text corresponding to the positive sample questions can be determined based on the analysis results. For example, if the recall text corresponding to the positive sample question adopts a keyword recall strategy, then when parsing historical questions, the keyword recall strategy can be used as a question parsing strategy, and keywords in the historical questions can be extracted and analyzed. Based on the analysis results, a question generation prompt text that more accurately matches the user's intention is generated; and through this automated sampling strategy, the need for manual annotation of prompt text can be reduced, thereby reducing development costs and accelerating the optimization process of the model.

[0154] In addition, in order to further improve the efficiency and accuracy of generating text questions corresponding to the target text, a large language model can be used to generate text questions corresponding to the target text. The specific implementation method is as follows:

[0155] The step of generating a prompt text according to the target text and the question, and generating a text question corresponding to the target text, comprises:

[0156] The target text and the question generation prompt text are input into the initial language processing model to obtain a text question corresponding to the target text.

[0157] In the language processing model training method provided in one or more embodiments of the present specification, the initial language processing model is used in combination with the constructed question generation prompt text to generate text questions for the target text. This can not only quickly generate text questions corresponding to the target text, but the model can also identify possible deviations or error tendencies based on the question generation prompt text, so that the initial language processing model can generate text questions that better match the target text, and it can be easier to obtain the required information or solutions later, thereby enhancing the user experience.

[0158] Step 204: Generate prompt text from the target text, the text question corresponding to the target text, and the text key points, and input them into an initial language processing model to obtain the text key points corresponding to the text question.

[0159] Specifically, after obtaining the target text and the text question corresponding to the target text according to the above implementation, the target text, the text question corresponding to the target text, and the text key points can be generated into a prompt text and input into an initial language processing model. The initial language processing model is used to obtain the text key points corresponding to the text question.

[0160] Step 206: construct text training data according to the target text, the text question corresponding to the target text, and the text key points corresponding to the text question.

[0161] After obtaining the text key points corresponding to the text question, text training data can be constructed according to the target text, the text question corresponding to the target text, and the text key points corresponding to the text question.

[0162] In practical applications, in order to ensure the quality and applicability of the text points corresponding to the text questions, the text points corresponding to the obtained text questions can be filtered. The specific implementation method is as follows:

[0163] The step of constructing text training data according to the target text, the text question corresponding to the target text, and the text key points corresponding to the text question includes:

[0164] Input the target text, the text question corresponding to the target text, the text key points corresponding to the text question, and the text key point filtering prompt text into the initial language processing model to obtain the filtered candidate text key points;

[0165] Text training data is constructed according to the candidate text main points, the text questions corresponding to the candidate text main points and the target text.

[0166] Specifically, the target text, the text question corresponding to the target text, the text key points corresponding to the text question, and the text key point filtering prompt text are input into the initial language processing model to obtain filtered candidate text key points. It can be understood that the target text, the text question corresponding to the target text, the text key points corresponding to the text question, and the text key point filtering prompt text are input into the initial language processing model, and the initial language processing model is used to filter and delete text key points containing redundant or repeated information, information irrelevant to the text question, sensitive information, etc., and then the text key points retained after filtering are used as candidate text key points; text training data is constructed according to the candidate text key points, the text question corresponding to the candidate text key points, and the target text.

[0167] In the language processing model training method provided in one or more embodiments of the present specification, the initial language processing model is used to filter the text key points corresponding to the text questions, which can ensure that the generated text key points have higher content quality, stronger relevance, better user experience, and are more secure and reliable; and greatly improve the data quality of the text training data constructed based on the filtered candidate text key points, the text questions corresponding to the candidate text key points, and the target text.

[0168] In specific implementation, in order to further improve the data quality of the constructed text training data, after filtering the text points according to the above method to obtain candidate text points, the quality of the candidate text points can also be evaluated, and the candidate text points can be further screened according to the evaluation results to obtain high-quality target text points, thereby further improving the data quality of the text training data constructed according to the target text points, the text questions corresponding to the target text points, and the target text. The specific implementation method is as follows:

[0169] The step of constructing text training data according to the candidate text main points, the text questions corresponding to the candidate text main points, and the target text includes:

[0170] Inputting the candidate text key points, the target text corresponding to the candidate text key points, and the text key point scoring prompt text into the initial language processing model to obtain the quality score of the candidate text key points;

[0171] Determine the candidate text points whose quality scores are greater than or equal to the preset quality scores as the target text points;

[0172] Text training data is constructed according to the target text main points, the text questions corresponding to the target text main points, and the target text.

[0173] Among them, the preset quality score can be set according to the actual application. For example, when the highest quality score is 10, in order to ensure the quality of the text training data, the preset quality score can be set to 8 points; or in order to ensure the data volume of the text training data and avoid too little training data, the preset quality score can be set to 7, etc.

[0174] Specifically, the candidate text main points, the target text corresponding to the candidate text main points, and the text main point scoring prompt text are first input into the initial language processing model, and the initial language processing model is used to obtain the quality score of the candidate text main points; then, the candidate text main points with a quality score greater than or equal to the preset quality score are determined as the target text main points; finally, according to the target text main points, the text questions corresponding to the target text main points and the target text, text training data is constructed to carry out subsequent model training of the initial language processing model.

[0175] In the language processing model training method provided in one or more embodiments of the present specification, an initial language processing model is used to obtain quality scores of candidate text main points, and candidate text main points with quality scores greater than or equal to preset quality scores are used as target text main points. The target text main points filtered by the quality scores can be regarded as high-quality text main points. Then, the text training data constructed based on the target text main points, the text questions corresponding to the target text main points, and the target text are of better quality and more representative. By training such data, the overfitting phenomenon of the model to specific data can be reduced, and the robustness of the model in the face of different scenarios and noise data can be increased, thereby improving the prediction accuracy of subsequent models.

[0176] 208: Train the initial language processing model according to the text training data to obtain a target language processing model.

[0177] Specifically, after constructing the text training data, the initial language processing model can be trained according to the constructed text training data to obtain a target language processing model, wherein the target language processing model is used to output the text key points corresponding to the text question when receiving the text question.

[0178] In practical applications, the language processing model after training can be used as the final target language processing model. In the case of multiple model iterations, the initial language processing model can be understood as the target language processing model after each iteration update.

[0179] The language processing model training method provided in one or more embodiments of the present specification utilizes a large language model in combination with a designed prompt to achieve semantic segmentation of text, accurate generation and filtering of text key points; through the powerful natural language understanding and generation capabilities of the large language model, it is possible to identify and extract high-value information from complex text data, significantly improving the efficiency and accuracy of information processing; specifically, the method first parses the initial text, segments it according to semantics, and reduces the number of text slices; then, through the designed prompt text combined with the user's historical question preferences, accurate text key points are generated for different topics and needs, relying on semantic segmentation of the text and the rewriting of the initial text by text key points, which not only improves the accuracy of key information extraction, but also enables the search process to more accurately recall the most relevant knowledge content; that is, the solution combining a large language model with a prompt and the user's historical query can promote the efficient use of information knowledge, thereby greatly improving the user's query efficiency and enhancing the user experience.

[0180] In practical applications, in order to ensure the performance, reliability, generalization ability and practical application effect of the target language processing model, the target language processing model can be tested so that the target language processing model can be subsequently processed according to the test results. The specific implementation method is as follows:

[0181] After training the initial language processing model according to the text training data to obtain the target language processing model, the method further includes:

[0182] Sampling the target text to obtain a test text;

[0183] According to the test text and the test text question corresponding to the test text, obtaining the test text key points corresponding to the test text question;

[0184] Constructing text test data according to the test text, the test text questions corresponding to the test text, and the test text key points corresponding to the test text questions;

[0185] According to the text test data, a model test is performed on the target language processing model to obtain a model test result.

[0186] Here, sampling the target text to obtain the test text can be understood as selecting a part of the target text as the test text.

[0187] The specific implementation of the test text questions corresponding to the test text, and obtaining the test text key points corresponding to the test text questions based on the test text and the test text questions corresponding to the test text, is the same as the specific implementation method of the text key points corresponding to the text questions of the target text in the above-mentioned embodiment, and will not be repeated here.

[0188] After obtaining the test text key points corresponding to the test text questions, text test data may be constructed according to the test text, the test text questions corresponding to the test text, and the test text key points corresponding to the test text questions.

[0189] According to the text test data, the target language processing model is subjected to model testing to obtain a model test result. This can be understood as inputting the test text problem in the text test data into the target language processing model to obtain the predicted text main points and the predicted text output by the target language processing model; and obtaining the model test result based on the matching relationship between the test text main points, the test text, and the predicted text main points and the predicted text corresponding to the test text problem. For example, in the case of a match, the model test result can be considered to be excellent, and in the case of a mismatch, the model test result can be considered to be poor, etc.

[0190] The language processing model training method provided in one or more embodiments of the present specification may, after training to obtain a target language processing model, sample from the target text to construct text test data in order to ensure the performance and reliability of the target language processing model in practical applications, and perform model testing on the target language processing model based on the text test data to verify whether the model can achieve the expected goal, and determine whether further iterative adjustments are needed based on the model test results, thereby continuously improving the model performance.

[0191] In addition, the specific execution process of the language processing model training method provided in one or more embodiments of this specification can be implemented on the server side. In order to improve the user interaction experience, the server side can respond to the model training request sent by the terminal device when receiving the model training request, execute the language processing model training method, and deploy the target language processing model obtained by executing the language processing model training method to the terminal device, or return the model call interface corresponding to the target language processing model to the terminal device. The specific implementation method is as follows:

[0192] The determining of the target text and the text problem corresponding to the target text includes:

[0193] In response to a model training request for the initial language processing model sent by a terminal device, determining the target text and a text question corresponding to the target text;

[0194] After training the initial language processing model according to the text training data to obtain the target language processing model, the method further includes:

[0195] The target language processing model or the model calling interface of the target language processing model is returned to the terminal device, so that the terminal device deploys the target language processing model or calls the target language processing model according to the model calling interface.

[0196] In order to facilitate an overall understanding of the language processing model training method provided by one or more embodiments of this specification, Figure 4 A flow chart of the overall processing process of a language processing model training method provided by an embodiment of the present specification is shown, which specifically includes the following steps.

[0197] Step 402: Obtain initial data.

[0198] Specifically, obtaining initial data can be understood as obtaining multiple initial texts in the target field.

[0199] Step 404: Obtain structured data of semantic segmentation.

[0200] Specifically, obtaining semantically segmented structured data can be understood as segmenting the initial data using preset text segmentation rules and then performing semantic segmentation through the initial language processing model to obtain multiple semantic text fragments, namely the target text in the above embodiment.

[0201] Step 406: Collect historical questions of some users.

[0202] Step 408: Generate a text question corresponding to the target text by using the large language model combined with the prompt text.

[0203] Specifically, the large language model is used in combination with the prompt text to generate the text question corresponding to the target text. This can be understood as first generating the prompt text based on the analysis of the historical questions collected from some users, and then using the initial language processing model to generate the prompt text in combination with the question to generate the text question corresponding to the target text.

[0204] Step 410: Output the text question corresponding to the target text.

[0205] Step 412: Utilize the large language model in combination with the prompt text to generate text key points corresponding to the text question.

[0206] Specifically, the initial language processing model is used to generate prompt text in combination with the text key points, and the text key points corresponding to the text question are generated. This can be understood as inputting the target text, the text question corresponding to the target text, and the prompt text generated in combination with the text key points into the initial language processing model, and using the initial language processing model to generate the text key points corresponding to the text question.

[0207] Step 414: Output text key points that are highly similar to the text question, highly generalizable, and highly relevant.

[0208] Step 416: fine-tune the model iteratively.

[0209] Specifically, fine-tuning and iterating the model can be understood as constructing text training data based on the target text, the text questions corresponding to the target text, and the text points corresponding to the text questions, and fine-tuning and iterating the initial language processing model (i.e., the large language model in step 408 and step 412), so as to obtain a target language processing model with better performance and further improve the model processing efficiency and effect.

[0210] In the language processing model training method provided in one or more embodiments of this specification, the text key point generation technology is mainly used in the recall stage in the search scenario. In this recall stage, the goal is to quickly identify and extract and query the most relevant information from a large number of documents or data sets. In order to achieve this goal, the text key point generation technology needs to be able to accurately extract the key information in the target text, which can summarize the content of the target text as comprehensively and concisely as possible; in order to perform well on some extreme problems, in the process of generating text key points, the text questions corresponding to the target text will be combined to generate them, and the potential information of the text or higher-level information similar to the text questions will be mined as much as possible to improve the accuracy in the recall scenario. In addition, in order to further improve the recall efficiency, the initial text will be processed by semantic segmentation to obtain a target text without redundant information and semantically complete and coherent, and the text key point generation technology mentioned above will be combined to generate text key points; in this way, the recall efficiency and accuracy of the subsequent target language processing model in the search scenario can be improved by semantic segmentation and combining the problem-guided generation of text key points.

[0211] The following combination Figure 5 , taking the application of the language processing model training method provided in this specification in the field of medical research as an example, the language processing model training method is further explained. Figure 5 This is a processing flow chart of a language processing model training method in the field of medical research provided by an embodiment of this specification, which specifically includes the following steps.

[0212] Step 502: Sample historical questions of some users.

[0213] Specifically, combined with the application scenarios in the field of medical research, sampling some users' historical problems can be understood as sampling some users' historical medical problems in the field of medical research.

[0214] Step 504: Generate question generation prompt text.

[0215] Specifically, in combination with the application scenario in the field of medical research, generating a prompt text for question generation can be understood as parsing the historical medical questions sampled in step 502, and constructing a prompt text for question generation that can subsequently generate medical questions using a large language model based on the parsing results.

[0216] Step 506: Acquire the target text from the semantic structured database.

[0217] The semantic structured database may be understood as the semantic database of the above-mentioned embodiment, and the target text may be understood as a plurality of semantic text segments stored in the semantic database after semantic segmentation and structural processing.

[0218] Specifically, in combination with the application scenario in the field of medical research, obtaining the target text from the semantically structured database can be understood as obtaining the target medical text from the semantically structured database corresponding to the field of medical research.

[0219] Step 508: Generate text questions corresponding to the target text.

[0220] Specifically, in combination with the application scenarios in the field of medical research, generating text questions corresponding to the target questions can be understood as using a large language model (i.e., the initial language processing model in the above embodiment) to generate prompt texts in combination with the problems in the field of medical research, and generating questions for the target medical text; that is, the target medical text and the prompt text for problem generation in the field of medical research are input into the large language model, and the large language model is used to generate questions for the target medical text, thereby generating medical text questions corresponding to the target medical text.

[0221] Step 510: Output the text question corresponding to the target text.

[0222] Specifically, combined with the application scenarios in the field of medical research, outputting text questions corresponding to the target text can be understood as a large language model outputting structured medical text questions corresponding to the target medical text.

[0223] Step 512: Post-process the structured data.

[0224] Specifically, combined with the application scenarios in the field of medical research, post-processing of structured data can be understood as data post-processing of the target medical text obtained from the semantically structured database corresponding to the field of medical research, such as null value filtering, multi-level subheading merging, etc., to improve the quality of subsequent text key point generation.

[0225] Step 514: Generate text key points.

[0226] Specifically, combining with the application scenarios in the field of medical research, generating text key points can be understood as generating prompt text from the target medical text, the medical text problem corresponding to the target medical text, and the text key points, inputting them into the large language model, and obtaining the medical text key points corresponding to the medical text problem.

[0227] Step 516: Filter text key points.

[0228] Specifically, combined with the application scenarios in the field of medical research, filtering text key points can be understood as inputting the target medical text, the medical text questions corresponding to the target medical text, the medical text key points corresponding to the medical text questions, and the text key point filtering prompt text into the initial language processing model to obtain the filtered candidate medical text key points.

[0229] Step 518: Score the main points of the text.

[0230] Specifically, in combination with the application scenarios in the field of medical research, scoring of text key points can be understood as inputting the candidate medical text key points, the target medical text corresponding to the candidate medical text key points, and the text key point scoring prompt text into the initial language processing model to obtain the quality score of the candidate medical text key points.

[0231] Step 520: Construct a training set.

[0232] Specifically, combining with the application scenarios in the field of medical research, constructing a training set can be understood as determining the candidate medical text points whose quality scores are greater than or equal to the preset quality scores as the target medical text points; constructing text training data, i.e., a training set, based on the target medical text points, the medical text questions corresponding to the target medical text points, and the target medical text.

[0233] Step 522: Fine-tune and iterate the large model based on the training set.

[0234] Step 524: Text key points are stored.

[0235] Specifically, combined with the application scenarios in the field of medical research, text key point storage can be understood as storing the key points of the target medical text in a medical text database for subsequent query recall.

[0236] Specifically, the execution order of step 524 is not affected by step 520 and step 522. In actual applications, step 524 and step 520 or step 522 can be executed in parallel.

[0237] Step 526: Construct a test data set.

[0238] Specifically, combining the application scenarios in the field of medical research, constructing a test dataset can be understood as sampling some target medical texts from the semantically structured database corresponding to the field of medical research to construct a test dataset, which is used to perform performance testing on the large language model after fine-tuning iterations, so as to select the large language model with better performance for subsequent applications.

[0239] In the language processing model training method provided in one or more embodiments of the present specification, the initial text can be firstly segmented by semantic segmentation technology to obtain the target text after semantic and structural processing, and then stored in the semantic structured database to ensure that when the key points of the text are extracted later, the large model can more accurately identify and extract the key information in the target text, ensure the semantic integrity and coherence of the key information, and avoid the problem of mixed content; then, by analyzing the historical questions, a prompt text for question generation is generated, and text questions are generated for the target text according to the prompt text generated by the question; thus, the generation of text key points in the target text can be guided in combination with the text questions, so that the text can be The generation of this key point can be more focused and specific, focusing on the text problem, and improving the ability to solve the situation in the search; that is, thanks to the guidance of the text problem, when generating the text key points, it can not only combine the text content, but also abstract the deep or high representation content outside the text, and dig out the potential information of the text and higher-level information; through the combination of semantic segmentation and text key point generation technology, this method has achieved a significant improvement in the recall accuracy of subsequent search scenarios, and enhanced the relevance to extreme problems. The generated text key points can better meet user needs and improve user experience, allowing users to find the required content from a large amount of data more quickly and accurately. The application of this language processing model training method in the field of medical research can greatly improve the work efficiency of medical researchers and help the smooth progress of medical research.

[0240] See also Figure 6 , Figure 6 A flowchart of a problem handling method provided by an embodiment of the present specification is shown, which specifically includes the following steps.

[0241] Step 602: Receive a target question, input the target question into a target language processing model, and obtain text key points corresponding to the target question.

[0242] Wherein, the target language processing model is obtained by training according to the language processing model training method of the above embodiment.

[0243] Step 604: Determine the target document corresponding to the text key point according to the text key point.

[0244] Step 606: Input the target question, the text key points, and the target document into the target language processing model to obtain a target answer corresponding to the target question.

[0245] The target problem can be understood as a problem in the target field. For example, when the target language processing model is trained by target text in the medical field, the target problem can be understood as a medical problem in the medical field.

[0246] Optionally, the receiving target question includes:

[0247] Receiving a target question sent by a terminal device, wherein the target question is sent by a user through a dialogue interaction interface of the terminal device;

[0248] After obtaining the target answer corresponding to the target question, the method further includes:

[0249] The target answer corresponding to the target question is returned to the terminal device, so that the terminal device displays it to the user through the dialogue interaction interface.

[0250] Specifically, when the question processing method is applied to a terminal device, the specific implementation of the question processing method is to receive a target question sent by a user through a user interaction interface of the terminal device, input the target question into a target language processing model deployed thereon, and obtain the text key points corresponding to the target question; then, based on the one-to-one correspondence between the text key points and the target document, search for the corresponding target document based on the text key points from a document database; then, generate a prompt text by combining the target question, text key points, and target document with the answer, input the prompt text into the target language processing model, obtain the target answer corresponding to the target question, and display it to the user through the user interaction interface.

[0251] In practical applications, when the problem solving method is applied to a terminal device, the target language processing model can be deployed on the terminal device, or the terminal device can call and use the model interface of the target language processing model provided by the server.

[0252] In another implementation, when the question processing method is applied to a server, the specific implementation of the question processing method is: receiving a target question sent by a terminal device, wherein the target question can be sent by a user to the terminal device through a user interaction interface of the terminal device and forwarded by the terminal device to the server; then inputting the target question into a trained target language processing model to obtain the text points corresponding to the target question; then, based on the one-to-one correspondence between the text points and the target document, searching for the corresponding target document based on the text points from a document database; then, combining the target question, text points, and target document with the answer to generate a prompt text, inputting the target language processing model to obtain the target answer corresponding to the target question, and returning the target answer to the terminal device so that the terminal device can display it to the user through the user interaction interface.

[0253] During specific implementation, the target language processing model can also output semantic text fragments corresponding to the main points of the text. According to the above embodiment, the semantic text fragment is obtained by semantic segmentation of the initial text and has an inevitable one-to-one correspondence with the initial text. Therefore, the initial text can also be directly determined based on the semantic text fragment corresponding to the main points of the text; finally, the prompt text can be generated based on the target question, text main points, semantic text fragments, target document (i.e., initial text) and the answer, and the target answer corresponding to the target question can be generated using the target language processing model.

[0254] The question processing method provided in one or more embodiments of the present specification, after obtaining the target language processing model according to the above-mentioned language processing model training method, can then, in a specific question and answer processing search scenario, that is, after receiving the target question, quickly and accurately obtain the text key points corresponding to the target question based on the target language processing model, thereby quickly and accurately acquiring the corresponding target document from a large number of documents based on the text key points, generating the target answer corresponding to the target question, and improving the user's question and answer experience.

[0255] See also Figure 7 , Figure 7 A flowchart of a method for handling medical problems provided by an embodiment of the present specification is shown, which specifically includes the following steps.

[0256] Step 702: Receive a target medical question, input the target medical question into a target language processing model, and obtain key points of a medical text corresponding to the target medical question.

[0257] Wherein, the target language processing model is obtained by training according to the above-mentioned language processing model training method.

[0258] Step 704: Determine the target medical document corresponding to the medical text key points according to the medical text key points.

[0259] Step 706: Input the target medical question, the key points of the medical text, and the target medical document into the target language processing model to obtain a target medical answer corresponding to the target medical question.

[0260] Specifically, the specific implementation steps of the medical problem processing method are the same as the specific implementation steps of the problem processing method in the above embodiment, and will not be repeated here.

[0261] The medical question processing method provided in one or more embodiments of the present specification, after obtaining the target language processing model according to the above-mentioned language processing model training method, can then, in a specific medical question and answer processing search scenario, that is, after receiving the target medical question, quickly and accurately obtain the text key points corresponding to the target question based on the target language processing model, thereby quickly and accurately acquiring the corresponding target medical document from a large number of documents based on the text key points, and generating the target medical answer corresponding to the target medical question, thereby improving the user's medical question and answer experience.

[0262] Corresponding to the above method embodiment, this specification also provides a language processing model training device embodiment, Figure 8 FIG. 1 is a schematic diagram showing the structure of a language processing model training device provided by an embodiment of the present specification. Figure 8 As shown, the device comprises:

[0263] A determination module 802 is configured to determine a target text and generate a text question corresponding to the target text;

[0264] The key point acquisition module 804 is configured to input the target text, the text question corresponding to the target text, and the prompt text generated by the text key points into the initial language processing model to obtain the text key points corresponding to the text question;

[0265] A training data construction module 806 is configured to construct text training data according to the target text, the text question corresponding to the target text, and the text key points corresponding to the text question;

[0266] The training module 808 is configured to train the initial language processing model according to the text training data to obtain a target language processing model.

[0267] Optionally, the determining module 802 is further configured to:

[0268] The collected historical questions are parsed, and prompt text is generated based on the question construction result. The prompt text is generated based on the target text and the question, and a text question corresponding to the target text is generated.

[0269] Optionally, the determining module 802 is further configured to:

[0270] Collect historical problems, perform multi-dimensional data analysis on the historical problems, and construct question generation prompt texts based on the multi-dimensional data analysis results.

[0271] Optionally, the determining module 802 is further configured to:

[0272] Collect historical questions, and obtain positive sample questions corresponding to the historical questions and recall texts corresponding to the positive sample questions;

[0273] Performing sample analysis on the positive sample questions and the recalled texts corresponding to the positive sample questions to obtain a question parsing strategy;

[0274] According to the problem analysis strategy, the historical problem is analyzed, and according to the analysis result, the question is constructed to generate prompt text.

[0275] Optionally, the determining module 802 is further configured to:

[0276] The target text and the question generation prompt text are input into the initial language processing model to obtain a text question corresponding to the target text.

[0277] Optionally, the determining module 802 is further configured to:

[0278] An initial text is obtained, and semantic segmentation is performed on the initial text to determine the target text.

[0279] Optionally, the determining module 802 is further configured to:

[0280] Inputting the initial text and the semantic segmentation prompt text into the initial language processing model to obtain a plurality of semantic text segments;

[0281] Text processing is performed on the multiple semantic text segments to determine the target text.

[0282] Optionally, the determining module 802 is further configured to:

[0283] Segmenting the initial text according to a preset text segmentation rule to obtain a plurality of initial text segments;

[0284] The multiple initial text segments and the semantic segmentation prompt text are input into the initial language processing model to obtain the multiple semantic text segments.

[0285] Optionally, the determining module 802 is further configured to:

[0286] Inputting the plurality of semantic text segments and text verification prompt text into the initial language processing model to obtain a verification result of each semantic text segment in the plurality of semantic text segments;

[0287] According to the verification results of the semantic text segments, the multiple semantic text segments are filtered to obtain filtered target semantic text segments;

[0288] The target semantic text segment is determined as the target text.

[0289] Optionally, the training data construction module 806 is further configured to:

[0290] Input the target text, the text question corresponding to the target text, the text key points corresponding to the text question, and the text key point filtering prompt text into the initial language processing model to obtain the filtered candidate text key points;

[0291] Text training data is constructed according to the candidate text main points, the text questions corresponding to the candidate text main points and the target text.

[0292] Optionally, the training data construction module 806 is further configured to:

[0293] Inputting the candidate text key points, the target text corresponding to the candidate text key points, and the text key point scoring prompt text into the initial language processing model to obtain the quality score of the candidate text key points;

[0294] Determine the candidate text points whose quality scores are greater than or equal to the preset quality scores as the target text points;

[0295] Text training data is constructed according to the target text main points, the text questions corresponding to the target text main points, and the target text.

[0296] Optionally, the device further comprises:

[0297] The test module is configured as follows:

[0298] Sampling the target text to obtain a test text;

[0299] According to the test text and the test text question corresponding to the test text, obtaining the test text key points corresponding to the test text question;

[0300] Constructing text test data according to the test text, the test text questions corresponding to the test text, and the test text key points corresponding to the test text questions;

[0301] According to the text test data, a model test is performed on the target language processing model to obtain a model test result.

[0302] Optionally, the determining module 802 is further configured to:

[0303] In response to a model training request for the initial language processing model sent by a terminal device, determining the target text and generating a text question corresponding to the target text;

[0304] The device further comprises:

[0305] The model returns a module that is configured as follows:

[0306] The target language processing model or the model calling interface of the target language processing model is returned to the terminal device, so that the terminal device deploys the target language processing model or calls the target language processing model according to the model calling interface.

[0307] The above is a schematic scheme of a language processing model training device of this embodiment. It should be noted that the technical scheme of the language processing model training device and the technical scheme of the language processing model training method described above belong to the same concept, and the details not described in detail in the technical scheme of the language processing model training device can be found in the description of the technical scheme of the language processing model training method described above.

[0308] Corresponding to the above method embodiment, this specification also provides a problem handling device embodiment, Fig. 9 FIG. 1 is a schematic diagram showing a structure of a problem handling device provided by an embodiment of the present specification. Fig. 9 As shown, the device comprises:

[0309] The key point acquisition module 902 is configured to receive a target question, input the target question into a target language processing model, and obtain the text key points corresponding to the target question, wherein the target language processing model is trained according to the above-mentioned language processing model training method;

[0310] The document determination module 904 is configured to determine the target document corresponding to the text key point according to the text key point;

[0311] The answer generation module 906 is configured to input the target question, the text key points, and the target document into the target language processing model to obtain a target answer corresponding to the target question.

[0312] Optionally, the key point obtaining module 902 is further configured to:

[0313] Receiving a target question sent by a terminal device, wherein the target question is sent by a user through a dialogue interaction interface of the terminal device;

[0314] The device further comprises:

[0315] The answer return module is configured as follows:

[0316] The target answer corresponding to the target question is returned to the terminal device, so that the terminal device displays it to the user through the dialogue interaction interface.

[0317] The above is a schematic scheme of a problem solving device of this embodiment. It should be noted that the technical scheme of the problem solving device and the technical scheme of the problem solving method described above are of the same concept, and the details not described in detail in the technical scheme of the problem solving device can be found in the description of the technical scheme of the problem solving method described above.

[0318] Corresponding to the above method embodiment, this specification also provides a medical problem processing device embodiment, Fig.10 FIG. 1 is a schematic diagram showing the structure of a medical problem processing device provided by an embodiment of the present specification. Fig.10 As shown, the device comprises:

[0319] The medical key points acquisition module 1002 is configured to receive a target medical question, input the target medical question into a target language processing model, and obtain the medical text key points corresponding to the target medical question, wherein the target language processing model is obtained by training according to the above-mentioned language processing model training method;

[0320] A medical document determination module 1004 is configured to determine a target medical document corresponding to the medical text key points according to the medical text key points;

[0321] The medical answer generation module 1006 is configured to input the target medical question, the key points of the medical text, and the target medical document into the target language processing model to obtain a target medical answer corresponding to the target medical question.

[0322] The above is a schematic scheme of a medical problem treatment device of this embodiment. It should be noted that the technical scheme of the medical problem treatment device and the technical scheme of the medical problem treatment method described above belong to the same concept, and the details not described in detail in the technical scheme of the medical problem treatment device can be referred to the description of the technical scheme of the medical problem treatment method described above.

[0323] Fig.11 The block diagram of a computing device 1100 according to one embodiment of the present specification is shown. The components of the computing device 1100 include but are not limited to a memory 1110 and a processor 1120. The processor 1120 is connected to the memory 1110 via a bus 1130, and a database 1150 is used to store data.

[0324] The computing device 1100 also includes an access device 1140 that enables the computing device 1100 to communicate via one or more networks 1160. Examples of these networks include a public switched telephone network (PSTN), a local area network (LAN), a wide area network (WAN), a personal area network (PAN), or a combination of communication networks such as the Internet. The access device 1140 may include one or more of any type of network interface, wired or wireless (e.g., a network interface card (NIC)), such as an IEEE 802.11 wireless local area network (WLAN) wireless interface, a World Wide Interoperability for Microwave Access (Wi-MAX) interface, an Ethernet interface, a Universal Serial Bus (USB) interface, a cellular network interface, a Bluetooth interface, a near field communication (NFC) interface, and the like.

[0325] In one embodiment of the present specification, the above components of the computing device 1100 and Fig.11 Other components not shown in the figure may also be connected to each other, for example, via a bus. It should be understood that Fig.11 The computing device structure block diagram shown is only for the purpose of illustration, and is not intended to limit the scope of this specification. Those skilled in the art can add or replace other components as needed.

[0326] The computing device 1100 may be any type of stationary or mobile computing device, including a mobile computer or mobile computing device (e.g., a tablet computer, a personal digital assistant, a laptop computer, a notebook computer, a netbook, etc.), a mobile phone (e.g., a smart phone), a wearable computing device (e.g., a smart watch, smart glasses, etc.), or other types of mobile devices, or a stationary computing device such as a desktop computer or a personal computer (PC). The computing device 1100 may also be a mobile or stationary server.

[0327] Among them, the memory 1110 is used to store computer programs / instructions, and the processor 1120 is used to execute the computer programs / instructions stored in the memory 1110. When the computer program / instructions are executed by the processor, the steps of the above-mentioned language processing model training method, problem processing method, or medical problem processing method are implemented.

[0328] The above is a schematic scheme of a computing device of this embodiment. It should be noted that the technical scheme of the computing device and the technical scheme of the above-mentioned language processing model training method, problem processing method, or medical problem processing method belong to the same concept, and the details not described in detail in the technical scheme of the computing device can be referred to the description of the technical scheme of the above-mentioned language processing model training method, problem processing method, or medical problem processing method.

[0329] An embodiment of the present specification also provides a computer-readable storage medium storing a computer program / instruction, which, when executed by a processor, implements the steps of the above-mentioned language processing model training method, problem processing method, or medical problem processing method.

[0330] The above is a schematic scheme of a computer-readable storage medium of this embodiment. It should be noted that the technical scheme of the storage medium and the technical scheme of the above-mentioned language processing model training method, problem processing method, or medical problem processing method belong to the same concept, and the details not described in detail in the technical scheme of the storage medium can be referred to the description of the technical scheme of the above-mentioned language processing model training method, problem processing method, or medical problem processing method.

[0331] An embodiment of the present specification also provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the above-mentioned language processing model training method, problem processing method, or medical problem processing method.

[0332] The above is a schematic scheme of a computer program product of this embodiment. It should be noted that the technical scheme of the computer program product and the technical scheme of the above-mentioned language processing model training method, problem processing method, or medical problem processing method belong to the same concept, and the details not described in detail in the technical scheme of the computer program product can be referred to the description of the technical scheme of the above-mentioned language processing model training method, problem processing method, or medical problem processing method.

[0333] The above is a description of a specific embodiment of the specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0334] The computer program / instruction includes computer program code, which may be in source code form, object code form, executable file or some intermediate form, etc. The computer readable medium may include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium, etc. It should be noted that the content contained in the computer readable medium may be appropriately increased or decreased according to the requirements of patent practice. For example, in some regions, according to patent practice, computer readable media do not include electric carrier signals and telecommunication signals.

[0335] It should be noted that, for the above-mentioned method embodiments, for the sake of simplicity of description, they are all expressed as a series of action combinations, but those skilled in the art should be aware that the embodiments of this specification are not limited by the order of the actions described, because according to the embodiments of this specification, some 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 embodiments of this specification.

[0336] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0337] The preferred embodiments of this specification disclosed above are only used to help explain this specification. The optional embodiments do not describe all the details in detail, nor do they limit the invention to only the specific implementation methods described. Obviously, many modifications and changes can be made according to the content of the embodiments of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the embodiments of this specification, so that technicians in the relevant technical field can well understand and use this specification. This specification is only limited by the claims and their full scope and equivalents.

Claims

1. A language processing model training method, comprising: Determine a target text, and generate text questions corresponding to the target text; The target text, the text question corresponding to the target text, and the text key points generate prompt text, and input them into the initial language processing model to obtain the text key points corresponding to the text question; Constructing text training data according to the target text, the text question corresponding to the target text, and the text key points corresponding to the text question; The initial language processing model is trained according to the text training data to obtain a target language processing model.

2. According to the language processing model training method of claim 1, the step of generating a text question corresponding to the target text comprises: The collected historical questions are parsed, and prompt text is generated based on the question construction result. The prompt text is generated based on the target text and the question, and a text question corresponding to the target text is generated.

3. The language processing model training method according to claim 2, wherein the step of parsing the collected historical questions and constructing the questions and generating prompt texts according to the parsing results comprises: Collect historical questions, and obtain positive sample questions corresponding to the historical questions and recall texts corresponding to the positive sample questions; Performing sample analysis on the positive sample questions and the recalled texts corresponding to the positive sample questions to obtain a question parsing strategy; According to the problem analysis strategy, the historical problem is analyzed, and according to the analysis result, the question is constructed to generate prompt text.

4. According to the language processing model training method according to any one of claims 1 to 3, the determining of the target text comprises: An initial text is obtained, and semantic segmentation is performed on the initial text to determine the target text.

5. According to the language processing model training method of claim 4, the semantic segmentation of the initial text to determine the target text comprises: Inputting the initial text and the semantic segmentation prompt text into the initial language processing model to obtain a plurality of semantic text segments; Text processing is performed on the multiple semantic text segments to determine the target text.

6. The language processing model training method according to claim 5, wherein the initial text and the semantic segmentation prompt text are input into the initial language processing model to obtain a plurality of semantic text segments, comprising: Segmenting the initial text according to a preset text segmentation rule to obtain a plurality of initial text segments; The multiple initial text segments and the semantic segmentation prompt text are input into the initial language processing model to obtain the multiple semantic text segments.

7. According to the language processing model training method of claim 5, the step of performing text processing on the plurality of semantic text segments to determine the target text comprises: Inputting the plurality of semantic text segments and text verification prompt text into the initial language processing model to obtain a verification result of each semantic text segment in the plurality of semantic text segments; According to the verification results of the semantic text segments, the multiple semantic text segments are filtered to obtain filtered target semantic text segments; The target semantic text segment is determined as the target text.

8. According to any one of claims 1 to 3, the language processing model training method, wherein the text training data is constructed based on the target text, the text question corresponding to the target text, and the text key points corresponding to the text question, comprising: Input the target text, the text question corresponding to the target text, the text key points corresponding to the text question, and the text key point filtering prompt text into the initial language processing model to obtain the filtered candidate text key points; Text training data is constructed according to the candidate text main points, the text questions corresponding to the candidate text main points and the target text.

9. The language processing model training method according to claim 8, wherein the text training data is constructed according to the candidate text main points, the text questions corresponding to the candidate text main points, and the target text, comprising: Inputting the candidate text key points, the target text corresponding to the candidate text key points, and the text key point scoring prompt text into the initial language processing model to obtain the quality score of the candidate text key points; Determine the candidate text points whose quality scores are greater than or equal to the preset quality scores as the target text points; Text training data is constructed according to the target text main points, the text questions corresponding to the target text main points, and the target text.

10. According to the language processing model training method according to any one of claims 1 to 3, the step of determining a target text and generating a text question corresponding to the target text comprises: In response to a model training request for the initial language processing model sent by a terminal device, determining the target text and generating a text question corresponding to the target text; After training the initial language processing model according to the text training data to obtain the target language processing model, the method further includes: The target language processing model or the model calling interface of the target language processing model is returned to the terminal device, so that the terminal device deploys the target language processing model or calls the target language processing model according to the model calling interface.

11. A problem handling method, comprising: Receiving a target question, inputting the target question into a target language processing model, and obtaining text key points corresponding to the target question, wherein the target language processing model is obtained by training according to the language processing model training method according to any one of claims 1 to 10; According to the main points of the text, determining the target document corresponding to the main points of the text; The target question, the text main points, and the target document are input into the target language processing model to obtain a target answer corresponding to the target question.

12. The problem handling method according to claim 11, wherein receiving the target problem comprises: Receiving a target question sent by a terminal device, wherein the target question is sent by a user through a dialogue interaction interface of the terminal device; After obtaining the target answer corresponding to the target question, the method further includes: The target answer corresponding to the target question is returned to the terminal device, so that the terminal device displays it to the user through the dialogue interaction interface.

13. A method for treating a medical problem, comprising: Receiving a target medical question, inputting the target medical question into a target language processing model, and obtaining key points of a medical text corresponding to the target medical question, wherein the target language processing model is obtained by training according to the language processing model training method according to any one of claims 1 to 10 above; According to the main points of the medical text, determining a target medical document corresponding to the main points of the medical text; The target medical question, the main points of the medical text, and the target medical document are input into the target language processing model to obtain a target medical answer corresponding to the target medical question.

14. A language processing model training device, comprising: A determination module is configured to determine a target text and generate a text question corresponding to the target text; A key point acquisition module is configured to input the target text, the text question corresponding to the target text, and the prompt text generated by the text key points into an initial language processing model to obtain the text key points corresponding to the text question; A training data construction module is configured to construct text training data according to the target text, the text question corresponding to the target text, and the text key points corresponding to the text question; The training module is configured to train the initial language processing model according to the text training data to obtain a target language processing model.

15. A problem handling device, comprising: A key point acquisition module is configured to receive a target question, input the target question into a target language processing model, and obtain text key points corresponding to the target question, wherein the target language processing model is obtained by training according to the language processing model training method according to any one of claims 1 to 10; A document determination module is configured to determine a target document corresponding to the text key point according to the text key point; The answer generation module is configured to input the target question, the text key points, and the target document into the target language processing model to obtain a target answer corresponding to the target question.

16. A medical problem treatment device, comprising: A medical key point acquisition module, configured to receive a target medical question, input the target medical question into a target language processing model, and obtain the medical text key points corresponding to the target medical question, wherein the target language processing model is obtained by training according to the language processing model training method according to any one of claims 1 to 10 above; A medical document determination module is configured to determine a target medical document corresponding to the medical text key points according to the medical text key points; The medical answer generation module is configured to input the target medical question, the main points of the medical text, and the target medical document into the target language processing model to obtain a target medical answer corresponding to the target medical question.

17. A computing device comprising: Memory and processor; The memory is used to store computer programs / instructions, and the processor is used to execute the computer programs / instructions. When the computer program / instructions are executed by the processor, the steps of the method described in any one of claims 1 to 13 are implemented.

18. A computer-readable storage medium storing a computer program / instruction, wherein the computer program / instruction, when executed by a processor, implements the steps of the method according to any one of claims 1 to 13.

19. A computer program product, comprising a computer program / instruction, which, when executed by a processor, implements the steps of the method according to any one of claims 1 to 13.

Images