Question reappearing model training method, question reappearing method and related device

Abstract

The invention provides a training method of a question reappearing model, a question reappearing method and a related device. An electronic device obtains a fine tuning sample of a pre-training model; wherein the fine tuning sample comprises a problem emphasis sample and an emphasis auxiliary sample in the automobile field, and the emphasis auxiliary sample is used for enhancing the understanding ability and emphasis ability of the pre-training model for the problem; and training the pre-training model through the fine tuning sample to obtain a problem emphasis model. Thus, during model training, not only is the problem emphasis sample in the automobile field used, but also the auxiliary sample for enhancing the understanding ability and emphasis ability of the pre-training model for the problem is introduced, so that the trained problem emphasis model can understand questions in the automobile field and improve the accuracy of the emphasis of the problem.

Description

Technical Field

[0001] This application relates to the field of machine learning, and more specifically, to a method for training a question restatement model, a question restatement method, and related devices. Background Art

[0002] Question restatement is a technology widely used in the IT field. Question restatement is to change the description of a question without changing its semantics, so that the question has a more clear goal and no ambiguity. It has wide applications in search scenarios and large model dialogue scenarios. For example, in the search field, the user's question may have grammar problems, may lack some sentence components or have spelling mistakes, which brings uncertainty to the understanding of the question. Through the question restatement technology, the user's question can be corrected, so as to provide a text with clear semantics and clear goal for subsequent processing.

[0003] Since the restated question can often better recall the content related to the question, this is crucial for improving the user experience. However, it is found in the practice process that for a specific professional field such as automobiles, the current restatement method has a poor restatement effect. Summary of the Invention

[0004] In order to overcome at least one deficiency in the prior art, this application provides a method for training a question restatement model, a question restatement method, and related devices, specifically including:

[0005] In the first aspect, this application provides a method for training a question restatement model, and the method includes:

[0006] Obtain fine-tuning samples of a pre-trained model, where the fine-tuning samples include question restatement samples and restatement auxiliary samples in the automotive field, and the restatement auxiliary samples are used to enhance the pre-trained model's understanding ability and restatement ability of questions;

[0007] Train the pre-trained model with the fine-tuning samples to obtain a question restatement model.

[0008] In the second aspect, this application provides a question restatement method, and the method includes:

[0009] Receive the original question input by the user;

[0010] Process the original question and a preset automotive brand tree through the question restatement model obtained by the method to obtain a restated question of the original question, where the automotive brand tree records a variety of automotive brands and the vehicle information of the vehicles already released by each automotive brand.

[0011] In the third aspect, this application provides a device for training a question restatement model, and the device includes:

[0012] A sample acquisition module for acquiring fine-tuning samples of a pre-trained model, where the fine-tuning samples include problem restatement samples and restatement auxiliary samples in the automotive field, and the restatement auxiliary samples are used to enhance the pre-trained model's ability to understand problems and restatement ability;

[0013] A model training module for training the pre-trained model with the fine-tuning samples to obtain a problem restatement model.

[0014] In a fourth aspect, the present application provides a problem restatement device, which includes:

[0015] A problem receiving module for receiving an original problem input by a user;

[0016] A problem restatement module for processing the original problem and a preset automotive brand tree through the problem restatement model obtained by the method to obtain a restatement problem of the original problem, where the automotive brand tree records multiple automotive brands and vehicle information of the vehicles released by each automotive brand.

[0017] In a fifth aspect, the present application provides a storage medium storing a computer program, which when executed by a processor, implements the training method of the problem restatement model or the problem restatement method.

[0018] In a sixth aspect, the present application provides an electronic device, which includes a processor and a memory, the memory stores a computer program, and when the computer program is executed by the processor, it implements the training method of the problem restatement model or the problem restatement method.

[0019] Compared with the prior art, the present application has the following beneficial effects:

[0020] In the training method of the problem restatement model, the problem restatement method and related devices provided by the present application, an electronic device acquires fine-tuning samples of a pre-trained model; among them, the fine-tuning samples include problem restatement samples and restatement auxiliary samples in the automotive field, and the restatement auxiliary samples are used to enhance the pre-trained model's ability to understand problems and restatement ability; and the pre-trained model is trained with the fine-tuning samples to obtain a problem restatement model. In this way, not only problem restatement samples in the automotive field are used during model training, but also auxiliary samples for enhancing the pre-trained model's ability to understand problems and restatement ability are introduced, so that the trained problem restatement model can understand questions in the automotive field and improve the accuracy of restating problems. Description of the Drawings

[0021] To more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the accompanying drawings required in the embodiments. It should be understood that the following drawings only show some embodiments of the present application and should not be regarded as limiting the scope. For those of ordinary skill in the art, without creative efforts, other related drawings can also be obtained based on these drawings.

[0022] Figure 1 It is a schematic flowchart of the training method of the problem restatement model provided by the embodiments of the present application;

[0023] Figure 2 It is one of the schematic flowcharts of the problem restatement method provided by the embodiments of the present application;

[0024] Figure 3 It is the second of the schematic flowcharts of the problem restatement method provided by the embodiments of the present application;

[0025] Figure 4 It is a schematic structural diagram of the training device of the problem restatement model provided by the embodiments of the present application;

[0026] Figure 5 It is a schematic structural diagram of the problem restatement device provided by the embodiments of the present application;

[0027] Figure 6 It is a schematic structural diagram of the electronic device provided by the embodiments of the present application. Detailed implementation manners

[0028] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are some, but not all, of the embodiments of the present application. Usually, the components of the embodiments of the present application described and shown in the accompanying drawings here can be arranged and designed in various different configurations.

[0029] Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the present application to be protected, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present application.

[0030] It should be noted that: similar reference numerals and letters denote similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings.

[0031] In the description of this application, it should be noted that the terms "first", "second", "third", etc. are only used for distinguishing descriptions and cannot be construed as indicating or implying relative importance. In addition, the terms "include", "comprise" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "including one..." does not exclude the existence of additional identical elements in the process, method, article or device including the said element.

[0032] Based on the above statement, as introduced in the background art, for a specific professional field such as automobiles, the current restatement method has a poor restatement effect.

[0033] During the research process, it was proposed to restate the problem through the rule method and the model method. The principles and disadvantages of these two methods will be described in detail below.

[0034] The rule method is a problem restatement method based on predefined rules. This method supplements missing sentence elements or pronouns in the problem by setting a series of fixed rules. For example, if a key noun is missing in a problem, the rule method can complete it according to the preset rules. The core of this method lies in using a fixed rule library, enabling the machine to automatically correct the problem according to these rules. The advantage of the rule method is that it is simple to implement, easy to program and maintain. Therefore, it can quickly supplement the missing parts of the problem and improve the clarity of the problem. However, due to relying on predefined rules, it is difficult for the rule method to comprehensively cover complex or variable problems. The supplemented problem may not conform to natural language habits and may not read smoothly. In addition, for problems in professional fields, the rule method may lack sufficient professional knowledge support, resulting in inaccurate restatement results.

[0035] The model method is another method of using a language model to restate problems. This method trains a large-scale language model to enable it to have the ability to understand and rewrite problems. The model method can learn a large amount of language data during the training process, thereby improving the accuracy and naturalness of problem restatement. After a large amount of training, the language model can generally generate natural and fluent problem restatements, and its performance on general problems is usually better than that of the rule method. However, the model method performs poorly in professional fields, especially in the case of insufficient professional knowledge. There may be defects in the model training process. For example, there may be ambiguity when dealing with professional terms. In addition, in some specific tasks, the restatement results generated by the model method may not be precise enough, leading to problems in subsequent processing links.

[0036] For example, in the automotive field, "Seal" actually refers to the name of a certain vehicle. However, when using a language model to restate a question, it is possible to recognize "Seal" as an animal. Subsequently, ambiguity is introduced into the restated question.

[0037] Based on the discovery of the above technical problems, the inventors have put forward the following technical solutions through creative labor to solve or improve the above problems. It should be noted that the defects existing in the above solutions in the prior art are the results obtained by the inventors through practice and careful research. Therefore, the process of discovering the above problems and the solutions proposed by the embodiments of the present application below for the above problems should be the contributions made by the inventors to the present application during the invention creation process, and should not be understood as the technical content known to those skilled in the art.

[0038] In view of the above problems, an embodiment of the present application (hereinafter simply referred to as this embodiment) provides a method for training a question restatement model. As Figure 1 shown, the method includes:

[0039] S1A, obtaining fine-tuning samples of a pre-trained model.

[0040] Among them, the fine-tuning samples include question restatement samples and restatement auxiliary samples in the automotive field. The restatement auxiliary samples are used to enhance the pre-trained model's ability to understand questions and its restatement ability.

[0041] S2A, training the pre-trained model with the fine-tuning samples to obtain a question restatement model.

[0042] In this way, not only question restatement samples in the automotive field are used during model training, but also auxiliary samples for enhancing the pre-trained model's ability to understand questions and its restatement ability are introduced, so that the trained question restatement model can understand questions in the automotive field and improve the accuracy of restating questions.

[0043] For the training method of the problem restatement model provided in this embodiment, the electronic device implementing this method can be, but is not limited to, a mobile terminal, a tablet computer, a laptop computer, a desktop computer, a server, etc., as long as it can provide sufficient computing power for model training. Among them, the server can be a single server or a server group. The server group can be centralized or distributed (for example, the server can be a distributed system). In some embodiments, the server can be local or remote relative to the user terminal. In some embodiments, the server can be implemented on a cloud platform; by way of example only, the cloud platform can include a private cloud, a public cloud, a hybrid cloud, a community cloud, a distributed cloud, an inter-cloud, a multi-cloud, etc., or any combination thereof. In some embodiments, the server can be implemented on an electronic device having one or more components.

[0044] To make the training method of the problem restatement model provided in this embodiment clearer, the following uses a server as the electronic device implementing this method to elaborate Figure 1 each step shown in detail. However, it should be understood that the operations in the flowchart may not be implemented in sequence, and steps without a logical context relationship can be reversed or implemented simultaneously. In addition, those skilled in the art can add one or more other operations to the flowchart or remove one or more operations from the flowchart under the guidance of the content of this application. Continuing to refer to Figure 1 , the method includes:

[0045] S1A, obtaining fine-tuning samples of the pre-trained model.

[0046] Among them, the fine-tuning samples include problem restatement samples and restatement auxiliary samples in the automotive field. The restatement auxiliary samples are used to enhance the pre-trained model's ability to understand and restate problems. As an alternative implementation, the restatement auxiliary samples include reading comprehension samples and pronoun recognition samples. Among them, the reading comprehension samples include long text content, and the label is the corresponding summarized content. The pronoun recognition samples include text content with pronouns, and the label is the entity object corresponding to the pronoun therein.

[0047] In this regard, it should be understood that during the research process, it was found that the restatement ability of the model is closely related to its ability to understand questions. To improve the model's restatement ability in the automotive field, it is necessary to ensure that the model can deeply understand the key information and background in the questions. For this purpose, multiple sets of automotive field data similar to reading comprehension were sorted out. The main purpose of this data is to help the model better understand complex texts in the automotive field. Specifically, the reading comprehension samples can include some articles in the automotive field, which can come from automotive magazines or automotive forums, and the summaries of these articles are given as labels during model training. In this way, the model can better understand the professional terms and background knowledge in the automotive field, and thus can more accurately grasp the key information of the question when restating the question, reducing ambiguity.

[0048] In addition, the research also found that users often use some pronouns when asking questions to refer to an entity mentioned in the context. For example, many questions contain pronouns (such as "he", "she", "it", etc.), and these pronouns may cause confusion. Therefore, in this embodiment, pronoun recognition samples are specially sorted out for training the model to recognize and process pronouns. The labels of these data mainly focus on identifying the pronouns in the questions and finding the specific entities referred to by these pronouns. For example, in the question "What is the price of the Magotan, and what is its wheelbase?", "he" is a pronoun that refers to the entity object "Magotan". The model needs to be able to recognize this pronoun and understand that the object referred to is the "Magotan" in the previous text. Through such training, the model can clearly point out the object referred to by the pronoun when restating the question, thus avoiding ambiguity and making the question expression clearer and more accurate. Therefore, eliminating ambiguity is a very crucial part when restating the question.

[0049] Among them, the above-mentioned pronoun samples can come from the automotive field or non-automotive fields. It can be understood that when training the question restatement model, not only the questions in the automotive field should be concerned, but also other types of questions that users may ask in non-automotive fields should be considered. To make the model more generalizable when dealing with various questions, pronoun samples are introduced from non-automotive fields.

[0050] For the above-mentioned question restatement samples, the server can obtain them through the following implementation methods:

[0051] S1A-1, obtain the user questions in the automotive field and the context of the user questions.

[0052] In this embodiment, "user question" refers to a specific question raised by a user in the automotive field to the system. For example, questions like "What is the fuel consumption of the Magotan?" usually appear on automotive forums or Q&A platforms. "Context" refers to the background information related to the user question or other questions in a series of questions. For example, before asking the above question, the user may have asked a question like "What is the engine type of the Magotan?" and the answer to this question. Then this series of questions constitutes the user's context information.

[0053] In this embodiment, the actual questions raised by users can be crawled from Q&A platforms or websites in the automotive field. These platforms can be automotive modules on professional automotive websites or large comprehensive websites because a large number of user questions and related discussions are gathered there. Then, collect the context information of the user questions, which includes the background of the questions raised by users and a possible series of related questions. For example, a series of questions raised by the user before and after asking a certain question can be regarded as part of the context. The collection of context information ensures that the model can better understand the background and relevance of the questions.

[0054] S1A-2, process the user question and the context of the user question through a third-party large model to generate a restated content of the user question.

[0055] Among them, the restated content includes an inference part and a question restatement part. In this regard, in this embodiment, a third-party large model (such as Deepseek, ChatGPT, etc.) can be used to process the user question and its context. The third-party large model will analyze the user question and its context to generate the restated content. The generated restated content not only contains a restated version of the user question but also includes the inference part, and these inference parts include the relevant background and intention of the question.

[0056] The so-called inference part means that in addition to generating the question restatement part, the third-party large model also needs to give the thinking process for generating the question restatement part. Specifically, the inference part mainly includes the following aspects:

[0057] First of all, the third-party large model needs to determine whether the current question is relevant to the previous user questions. For example, if the user asks "What is the fuel consumption of the Magotan?" and has previously asked "What is the engine type of the Magotan?", the third-party large model needs to understand the relevance between these two questions. This relevance can help the third-party large model better understand the background and purpose of the user question.

[0058] Secondly, the third-party large model also needs to identify the car brand, series, or model involved in the current question. For example, when dealing with the question "What is the fuel consumption of a Magotan?", the third-party large model needs to identify that "Magotan" is a specific car brand and model. Identifying these entities helps the third-party large model understand the specific object in the question more accurately.

[0059] Finally, the third-party large model needs to decide whether to restate the question. Some questions are already very clear and do not require restatement. For example, the question "What is the fuel consumption of a 2020 Toyota Corolla 1.8L automatic model?" already contains elements such as the brand, time, and model, and is clear enough. However, for those ambiguous questions, the third-party large model needs to generate a clearer restated version. In this way, the third-party large model can generate complete information that includes not only the question restatement but also background reasoning.

[0060] Exemplarily, assume the user's question is "What is the fuel consumption of a Magotan?" The third-party large model may reason about this question as follows:

[0061] "The user asks 'What is the fuel consumption of a Magotan'. This question is related to the previous historical question 'What is the engine type of a Magotan' because the engine type may affect fuel consumption. The user may want to know the fuel consumption performance of a Magotan (specific model) under different driving conditions, especially considering its engine type (such as turbocharged or naturally aspirated). Therefore, it is necessary to restate the question to include more details and background information, such as requesting detailed fuel consumption data and explaining how different engine types affect fuel consumption."

[0062] Finally, based on the above reasoning content, the third-party large model restates the question as:

[0063] "The user wants to know the fuel consumption performance of a Magotan (specific model) under different driving conditions, especially considering its engine type (such as turbocharged or naturally aspirated). Please provide detailed fuel consumption data and explain how different engine types affect fuel consumption."

[0064] Based on the above description of the user's question and the restated content, step S1 also includes:

[0065] S1A-3, after hiding some content in the restated content, use it together with the user's question as a question restatement sample.

[0066] In this embodiment, the server can hide part of the restated question or the reasoning part, enabling the model to predict the hidden part. It can be understood that part of the restated content or certain details in the reasoning process can be selectively hidden, so that the model can predict the hidden part based on the remaining unhidden part during model training. In this way, through multiple such question restatement samples, not only can the model's question restatement ability be trained, but also the model's reasoning ability can be trained. For example, for a "user question", the third-party large model can obtain an "inference part" and a "question restatement part". Assuming that the "inference part" is hidden, and the remaining "user question" and "question restatement part" are input into the model to train the model to generate the "inference part", so that the model has a certain reasoning ability.

[0067] S2A. Train the pre-trained model with fine-tuning samples to obtain a question restatement model.

[0068] The above pre-trained model refers to a large language model (LLM) pre-trained on a large-scale dataset. This embodiment does not specifically limit which large language model to use. For example, the base models of the Qwen series, DeepSeek-V3, Baichuan-Base, etc. Such models have learned rich feature representations and generalization abilities through training on a large amount of data. For example, the pre-trained model can be a pre-trained language model. Since the pre-trained model is a model pre-trained on a large-scale unlabeled data and has a certain basic language understanding ability, it can be fine-tuned in a supervised manner through specific task samples to enable it to better solve specific tasks. In this process, this embodiment uses three types of samples: question restatement samples, reading comprehension samples, and pronoun recognition samples. As an alternative implementation, step S2A may include:

[0069] S2A-1. Train the pre-trained model with question restatement samples, reading comprehension samples, and pronoun recognition samples respectively to obtain an initial model.

[0070] In this regard, the server can fine-tune the pre-trained model through these three types of samples, and each sample type corresponds to a specific task. Among them, the question restatement task requires the model to learn to restate questions, the reading comprehension task requires the model to understand and answer questions, and the pronoun recognition task requires the model to accurately identify pronouns in the text and their referents. Through this step-by-step training, the model gradually masters the basic skills in these tasks and forms an initial model.

[0071] S2A-2. Perform mixed training on the initial model with question restatement samples, reading comprehension samples, and pronoun recognition samples to obtain a question repetition model.

[0072] Next, to further improve the performance of the model on specific tasks, the initial model is continuously trained in a mixed manner. Mixed training means that in the same training process, these three different types of task samples are used simultaneously, and the training losses of the three different types of task samples are weighted and summed as the comprehensive model loss to adjust the model's parameters. In this way, the model not only continues to consolidate the skills learned previously but also can establish connections between these different tasks and better understand the complexity of language. Finally, the model after mixed training performs more excellently when dealing with problem repetition tasks.

[0073] Based on the problem restatement model obtained from the above embodiments, this embodiment also provides a problem restatement method. As Figure 2 shown, the method includes:

[0074] S1B, receiving the original problem input by the user.

[0075] S2B, processing the original problem and the preset car brand tree through the problem restatement model to obtain the restated problem of the original problem.

[0076] Among them, the car brand tree records multiple car brands and the vehicle information of the vehicles already released for each car brand.

[0077] In this way, when the problem restatement model optimizes the original problem, taking the car brand tree as embedded prior knowledge can help the model better understand the relevance and classification structure between car brands, provide important support for the problem restatement task, and enable the model to generate more targeted and diverse restated problems.

[0078] For the problem restatement method provided in this embodiment, the electronic device implementing this method can be, but is not limited to, a mobile terminal, a tablet computer, a laptop computer, a desktop computer, a server, etc. Among them, the server can be a single server or a server group. The server group can be centralized or distributed (for example, the server can be a distributed system). In some embodiments, the server can be local or remote relative to the user terminal. In some embodiments, the server can be implemented on a cloud platform; only as an example, the cloud platform can include a private cloud, a public cloud, a hybrid cloud, a community cloud, a distributed cloud, an inter-cloud, a multi-cloud, etc., or any combination thereof. In some embodiments, the server can be implemented on an electronic device with one or more components.

[0079] To make the problem restatement method provided in this embodiment clearer, the following continues to use the server as the electronic device implementing this method to Figure 2Each step shown will be elaborated in detail. However, it should be understood that the operations of the flowchart may not be implemented in sequence, and steps without logical context relationships can be reversed or implemented simultaneously. In addition, those skilled in the art can add one or more other operations to the flowchart or remove one or more operations from the flowchart under the guidance of the content of this application. Continuing to refer to Figure 2 , the method includes:

[0080] S1B. Receive the original question input by the user.

[0081] Exemplarily, assume that the user is using an online Q&A platform in the automotive field. The user encounters a problem: "I want to buy a Corolla as a commuting car. What is its fuel consumption?" At this time, the user enters this sentence in the search box of the platform, which constitutes an original question. After the server receives the input original question, the next task is to optimize this question through the trained question reformulation model to ensure that the expression of the question is clearer and more accurate.

[0082] S2B. Process the original question and the preset automotive brand tree through the question reformulation model to obtain the reformulated question of the original question.

[0083] Among them, the automotive brand tree records multiple automotive brands and the vehicle information of the vehicles already released for each automotive brand. In other words, the "automotive brand tree" is a data structure used to record and organize the detailed information of various automotive brands and the vehicles they have released. Similar to a tree diagram, the root node is the automotive brand, and the branches are the various vehicle series and models under that brand, thus providing prior knowledge in the automotive field for the question reformulation model.

[0084] It was also found during the research that current question reformulation methods usually directly submit the reformulated question after optimizing the original question to other large models for processing, or feedback the reformulated question to the user for confirmation. However, both of these methods have obvious limitations. Among them, the former, due to the lack of a strict verification mechanism for the reformulated question, cannot effectively guarantee the accuracy and reliability of the reformulated question. This may lead to deviations or misunderstandings in subsequent processing links, affecting the quality of the final result. The latter overly relies on the user's questioning ability and understanding depth of the question. If the user does not have sufficient professional knowledge or expression ability, it may lead to the reformulated question not being precise enough, thereby affecting the efficiency and effect of the entire problem-solving process.

[0085] During the research process, it was also found that when asking questions about a vehicle, if accurate information is to be obtained, the complete elements required for the vehicle problem need to be included, and the complete elements need to include "vehicle brand, model, production year, engine type, specific requirements", etc. Therefore, this embodiment can use the pre-provided complete elements to verify the restated problem. Based on this inventive concept, as Figure 3 shown, on the basis of Figure 2 , the problem restatement method provided in this embodiment further includes:

[0086] S3B, obtaining the existing elements in the restated problem.

[0087] For example, the server can continue to call the problem restatement model to process the restated problem and extract the existing elements in the restated problem from it.

[0088] S4B, if there are missing elements in the restated problem compared with the preset complete elements, the information corresponding to the missing elements in the vehicle brand tree is used as the missing information of the restated problem.

[0089] In this regard, in order to reduce the number of missing information, the server can obtain the existing information corresponding to the existing elements from the restated problem according to the existing elements, and use the existing information to screen the vehicle brand tree to match the target vehicle with the existing information from it; and obtain the missing information corresponding to the missing elements from the vehicle information of the target vehicle.

[0090] S5B, supplementing the missing information into the restated problem to obtain the secondary restated problem of the original problem.

[0091] Exemplarily, assume that the restated problem is "What is the fuel consumption of the Toyota Corolla 1.8L hybrid version?" After analysis by the server, it is found that the existing elements among them include "vehicle brand, model, engine type, specific requirements". Compared with the above complete elements, the missing element is "production year". At this time, the server uses "Corolla" and "1.8L hybrid version" to screen the vehicle brand tree to find the target vehicles that meet the two conditions of "Corolla" and "1.8L hybrid version" at the same time. Assume that there are 3 target vehicles here, and the production years of these target vehicles are 2022, 2021, and 2019 respectively. The server supplements these year information as the missing information into "What is the fuel consumption of the Toyota Corolla 1.8L hybrid version?" to obtain the secondary restated problem "What is the fuel consumption of the Toyota Corolla 1.8L hybrid version? The years are: 2022, 2021, 2019". Finally, the server submits the secondary restated problem to other pre-trained large models (such as the automotive Q&A model) for processing to obtain the answer to this problem.

[0092] Based on the same inventive concept as the training method of the problem restatement model provided in this embodiment, this embodiment also provides a training device for the problem restatement model, which includes at least one software function module that can be stored in a memory or fixed in an electronic device in the form of software. The processor in the electronic device is used to execute the executable module stored in the memory. For example, the software function module and computer program included in the device. Please refer to Figure 4 , functionally speaking, the device may include:

[0093] The sample acquisition module 11A is used to acquire fine-tuning samples of the pre-trained model, wherein the fine-tuning samples include problem restatement samples and restatement auxiliary samples in the automotive field, and the restatement auxiliary samples are used to enhance the pre-trained model's ability to understand and restate problems;

[0094] The model training module 12A is used to train the pre-trained model by fine-tuning samples to obtain a question restatement model.

[0095] In this embodiment, the sample acquisition module 11A is used to implement Figure 1 In step S1A, the model training module 12A is used to implement Figure 1 Therefore, for the detailed description of each of the above modules, please refer to the specific implementation of the corresponding steps, which will not be described in detail in this embodiment.

[0096] In addition, since the invention concept is the same as that of the training method of the problem restatement model, the training device of the problem restatement model can also implement other steps or sub-steps of the method through the above modules.

[0097] Optionally, the model training module 12A is further specifically used for:

[0098] The pre-training model is trained separately through question restatement samples, reading comprehension samples, and pronoun recognition samples to obtain the initial model;

[0099] The initial model is trained through mixed training of problem restatement samples, reading comprehension samples and pronoun recognition samples to obtain the problem repetition model.

[0100] Optionally, the sample acquisition module 11A is further specifically configured to:

[0101] Obtain user questions in the automotive field and the context of their questions;

[0102] Processing the user's question and the context of the user's question through a third-party big model to generate a restatement of the user's question, wherein the restatement includes a reasoning part and a restatement of the question;

[0103] After hiding some content in the restated content, it is used as a question restatement sample together with the user's question.

[0104] In addition, this embodiment also provides a question restatement device. As Figure 5 shown, the device includes:

[0105] A question receiving module 11B, configured to receive the original question input by the user;

[0106] A question restatement module 12B, configured to process the original question and a preset car brand tree through a question restatement model to obtain a restated question of the original question, where the car brand tree records multiple car brands and vehicle information of each released vehicle of each car brand.

[0107] In this embodiment, the question receiving module 11B is used to implement Figure 2 step S1B in Figure 2 and the question restatement module 12B is used to implement

[0108] step S2B in

[0109] Therefore, for the detailed description of the above modules, reference can be made to the specific implementation manners of the corresponding steps, and this embodiment will not elaborate on this.

[0110] Optionally, the question restatement module 12B is further configured to:

[0111] Obtain the existing elements in the restated question;

[0112] If there are missing elements in the restated question compared with the preset complete elements, then use the information corresponding to the missing elements in the car brand tree as the missing information of the restated question;

[0113] In addition, in each embodiment of the present application, the functional modules can be integrated together to form an independent part, or each module can exist alone, or two or more modules can be integrated to form an independent part.

[0114] It should also be understood that if the above embodiments are implemented in the form of software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present application.

[0115] Therefore, this embodiment also provides a storage medium that stores a computer program. When the computer program is executed by a processor, it implements the problem restatement method provided in this embodiment. Among them, the storage medium can be various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk, or an optical disc.

[0116] An electronic device for implementing the training method of the problem restatement model or the problem restatement method provided in this embodiment. As Figure 6 shown, the electronic device may include a processor 22 and a memory 21. And, the memory 21 stores a computer program. The processor reads and executes the computer program corresponding to the above embodiments in the memory 21 to implement the training method of the problem restatement model or the problem restatement method provided in this embodiment.

[0117] Continue to refer to Figure 6 , the electronic device also includes a communication unit 23. Each element of the memory 21, the processor 22, and the communication unit 23 is directly or indirectly electrically connected through a system bus 24 to achieve data transmission or interaction.

[0118] Among them, the memory 21 can be an information recording device based on any electronic, magnetic, optical, or other physical principles for recording execution instructions, data, etc. In some embodiments, the memory 21 can be, but is not limited to, a volatile memory, a non-volatile memory, a storage drive, etc.

[0119] In some embodiments, the volatile memory may be a Random Access Memory (RAM); in some embodiments, the non-volatile memory may be a Read Only Memory (ROM), a Programmable Read-Only Memory (PROM), an Erasable Programmable Read-Only Memory (EPROM), an Electric Erasable Programmable Read-Only Memory (EEPROM), a flash memory, etc.; in some embodiments, the storage drive may be a disk drive, a solid state drive, any type of storage disk (such as an optical disk, a DVD, etc.), or a similar storage medium, or a combination thereof, etc.

[0120] The communication unit 23 is configured to transmit and receive data via a network. In some embodiments, the network may include a wired network, a wireless network, an optical fiber network, a telecommunication network, an intranet, the Internet, a Local Area Network (LAN), a Wide Area Network (WAN), a Wireless Local Area Networks (WLAN), a Metropolitan Area Network (MAN), a Wide Area Network (WAN), a Public Switched Telephone Network (PSTN), a Bluetooth network, a ZigBee network, or a Near Field Communication (NFC) network, etc., or any combination thereof. In some embodiments, the network may include one or more network access points. For example, the network may include a wired or wireless network access point, such as a base station and / or a network switching node, and one or more components of the service request processing system may be connected to the network via the access point to exchange data and / or information.

[0121] The processor 22 may be an integrated circuit chip with signal processing capabilities, and the processor may include one or more processing cores (e.g., a single-core processor or a multi-core processor). By way of example only, the aforementioned processor may include a Central Processing Unit (CPU), an Application Specific Integrated Circuit (ASIC), an Application Specific Instruction-set Processor (ASIP), a Graphics Processing Unit (GPU), a Physics Processing Unit (PPU), a Digital Signal Processor (DSP), a Field Programmable Gate Array (FPGA), a Programmable Logic Device (PLD), a controller, a microcontroller unit, a Reduced Instruction Set Computing (RISC), or a microprocessor, etc., or any combination thereof.

[0122] It can be understood that Figure 6 The structure shown is only illustrative. The electronic device may also have more or fewer components than Figure 6 shown, or have a different configuration from Figure 6 that shown. Figure 6 Each of the components shown may be implemented using hardware, software, or a combination thereof.

[0123] It should be understood that the devices and methods disclosed in the above embodiments can also be implemented in other ways. The device embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architectures, functions, and operations of devices, methods, and computer program products according to multiple embodiments of the present application. In this regard, each block in the flowchart or block diagram may represent a module, a program segment, or a part of code, and the module, program segment, or part of code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than that marked in the accompanying drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, and the combination of blocks in the block diagram and / or flowchart, can be implemented by a dedicated hardware-based system that performs the specified functions or actions, or can be implemented by a combination of dedicated hardware and computer instructions.

[0124] As described above, these are only various embodiments of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed in the present application can easily think of changes or substitutions, which should all be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A training method for a problem restatement model, characterized in that: The method comprises: Obtain fine-tuning samples of the pre-trained model, wherein the fine-tuning samples include problem restatement samples and restatement auxiliary samples in the automotive field, and the restatement auxiliary samples are used to enhance the pre-trained model's ability to understand and restate problems; The pre-trained model is trained using the fine-tuning samples to obtain a question restatement model.

2. The method for training a problem restatement model according to claim 1, characterized in that: The restatement-aided samples include a reading comprehension sample and a pronoun recognition sample.

3. The method for training the problem restatement model according to claim 2, characterized in that: The pre-trained model is trained by the fine-tuning sample to obtain a question restatement model, including: The pre-training model is trained respectively by the problem restatement sample, the reading comprehension sample and the pronoun recognition sample to obtain an initial model; The initial model is mixedly trained by using the problem restatement samples, the reading comprehension samples and the pronoun recognition samples to obtain the problem repetition model.

4. The method for training the problem restatement model according to claim 1, characterized in that: Get sample restatements of the pre-trained model, including: Obtain user questions in the automotive field and contexts of the user questions; Processing the user's question and the context of the user's question through a third-party big model to generate a restatement of the user's question, wherein the restatement includes an inference part and a question restatement part; After part of the restated content is hidden, it is used together with the user's question as the question restatement sample.

5. A method for restating a problem, characterized in that: The method comprises: Receive the original question from the user; The original question and a preset car brand tree are processed by the question restatement model obtained by the method described in any one of claims 1 to 4 to obtain a restated question of the original question, wherein the car brand tree records multiple car brands and vehicle information of each of the car brands that have been released.

6. The problem restatement method according to claim 5, characterized in that: The method further comprises: Obtaining existing elements of the restated problem; If there are missing elements in the restated question compared with the preset complete elements, the information corresponding to the missing elements in the automobile brand tree is used as the missing information of the restated question; The missing information is added to the restated question to obtain a secondary restated question of the original question.

7. A training device for a problem restatement model, characterized in that: The device comprises: A sample acquisition module, used to acquire fine-tuning samples of the pre-trained model, wherein the fine-tuning samples include problem restatement samples and restatement auxiliary samples in the automotive field, and the restatement auxiliary samples are used to enhance the pre-trained model's ability to understand and restate problems; The model training module is used to train the pre-trained model through the fine-tuning samples to obtain a problem restatement model.

8. A question restatement device, characterized in that: The device comprises: A question receiving module, used for receiving the original question input by the user; A question restatement module is used to process the original question and a preset car brand tree through the question restatement model obtained by the method described in any one of claims 1 to 5 to obtain a restated question of the original question, wherein the car brand tree records multiple car brands and vehicle information of each of the car brands that have been released.

9. A storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by a processor, it implements the training method of the problem restatement model described in any one of claims 1-4 or the problem restatement method described in any one of claims 5-6.

10. An electronic device, characterized in that: The electronic device includes a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, it implements the training method of the problem restatement model described in any one of claims 1-4 or the problem restatement method described in any one of claims 5-6.

Images