Intelligent question-answering method and system for automotive electronic manual

Abstract

This application belongs to the field of intelligent vehicle technology, and particularly relates to an intelligent question-answering method and system for automotive electronic manuals. The question-answering method includes: S1, rewriting the user's question into a standardized question based on the vehicle model, and then labeling the domain categories associated with the current standardized question; S2, rewriting the current standardized question into standardized sub-requirements of the associated domain categories, and then obtaining the search results corresponding to each standardized sub-requirement based on the automotive electronic manual, and aggregating them into a domain knowledge set for the current standardized question; S3, feeding the current user question and the domain knowledge set into a large language model, and the large language model outputs the answer corresponding to the current user question. This application can efficiently and accurately output answers to user questions based on automotive electronic manuals.

Description

Technical Field

[0001] This application belongs to the field of intelligent vehicle technology, and in particular relates to an intelligent question-and-answer method and system for automotive electronic manuals. Background Technology

[0002] The car manual basically covers the problems that users will encounter throughout the entire life cycle of the vehicle. Therefore, when users encounter problems while using the vehicle, they will generally choose to look up the answers in the car manual.

[0003] As automotive electronics become increasingly sophisticated, car manuals have evolved into electronic manuals, with their content becoming more complex, covering technical parameters, troubleshooting, maintenance procedures, and operating guidelines. To facilitate users' quick access to answers (and subsequent actions based on those answers), the current mainstream approach is to employ a search-enhanced generation architecture to transform user questions into requirements, and then use those requirements to provide relevant content from the electronic car manual.

[0004] Because the questions are raised by users, they are highly colloquial and personalized. Sometimes, two completely different needs can appear very similar when expressed in natural language. If the question is transformed into an incorrect need, it will directly lead to incorrect feedback. If all possible needs for a question are output, it will significantly increase the time it takes for users to obtain truly effective information, which is not conducive to users solving problems quickly. Summary of the Invention

[0005] The purpose of this application is to overcome the shortcomings of the prior art and provide an intelligent question-answering method for automotive electronic manuals, which can efficiently and accurately output answers to user questions based on automotive electronic manuals.

[0006] To achieve the above objectives, this application adopts the following technical solution: A smart question-answering method for automotive electronic manuals includes: S1, after rewriting user questions into standardized questions based on vehicle models, labels the domain categories associated with the current standardized questions; S2, rewrite the current standardization problem into standardization sub-requirements of related domain categories, and then collect the search results corresponding to each standardization sub-requirement based on the automotive electronic manual to form a domain knowledge set for the current standardization problem; S3 feeds the current user question and the set of domain knowledge into the large language model, and the large language model outputs the answer corresponding to the current user question.

[0007] Preferably, S1 further includes the following sub-steps: S11: Convert the acquired user questions into text format; at the same time, retrieve the corresponding vehicle terminology database based on the vehicle model code; S12, based on the corresponding vehicle terminology library, standardized questions are obtained by rewriting vehicle terminology in user questions; S13, the current standardization problem is fed into the domain category labeling model, and the domain category labeling model labels the related domain categories of the current standardization problem; the domain categories include parameter query category, operation guide category, maintenance category, fault Q&A category, and timeliness information category.

[0008] Preferably, S12 is replaced with S12´: S12´, based on the corresponding vehicle terminology library, after rewriting the vehicle terminology in the user question, and then based on the general vehicle terminology library, the part of the user question that has not been rewritten with general terminology is rewritten with general terminology to obtain the standardized question.

[0009] Preferably, S13' is used instead of S13: S13' sends the current standardized problem and user interaction history into the domain category labeling model, and the domain category labeling model labels the associated domain category of the current standardized problem; the domain categories include parameter query category, operation guide category, maintenance category, fault Q&A category, and timeliness information category.

[0010] Preferably, the current standardization problem is rewritten into standardization sub-requirements of related domain categories, and the search results corresponding to each standardization sub-requirement are obtained based on the automotive electronic manual. This also includes the following: For the domain categories of parameter query and maintenance, the current standardization problem is rewritten into a first standardization sub-requirement that conforms to the structured query statement of SQL syntax; then, the first standardization sub-requirement is used to obtain the corresponding first search result based on the relational database containing parameter query and maintenance information in the automotive electronic manual. For the domain category of fault question and answer, the current standardized question is rewritten as a second standardized sub-requirement with a <entity-relationship-attribute value> triple probe structure; then, the second standardized sub-requirement is used to obtain the corresponding second search results based on the knowledge graph containing fault question and answer information in the automotive electronic manual. For the category of operation guides, after removing the noise of modal words and function words in the current standardization problem, the text is rewritten into an instructional phrase and then vectorized to obtain the third standardization sub-requirement; then, the third standardization sub-requirement is used to obtain the corresponding third search results based on the text containing operation guide information in the car electronic manual. For time-sensitive information categories, extract entities and time-sensitive keywords from the current standardization problem, then expand the synonyms of entities and time-sensitive information, remove stop words, and construct Boolean query logic to obtain the fourth standardization sub-requirement; then use the fourth standardization sub-requirement to obtain the corresponding fourth search results based on the search engine.

[0011] Preferably, in S3, the method further includes: feeding the current vehicle background information into the large language model.

[0012] Preferably, obtaining vehicle background information also includes the following: Step 1: Obtain user profiles, real-time multi-modal user information, vehicle status information, and vehicle driving environment information; Step 2: Use the first LLM model based on user profile and real-time multimodal user information to output personalized user tags; at the same time, use the second LLM model based on vehicle status information and vehicle driving environment information to output vehicle scene tags. Step 3: Then, use the third LLM model to merge the user's personalized tags and vehicle scene tags into vehicle background information.

[0013] This application also provides an intelligent question-answering system for automotive electronic manuals, including: a standardized question acquisition module, a domain category annotation module, a domain knowledge set acquisition module, and an answer acquisition module; the standardized question acquisition module is used to rewrite user questions into standardized questions based on vehicle models and then send them to the domain category annotation module; simultaneously, the standardized question acquisition module also sends user questions to the answer acquisition module; the domain category annotation module is used to annotate the domain categories associated with the standardized questions and then send them to the domain knowledge set acquisition module; the domain knowledge set acquisition module is used to rewrite the current standardized question into standardized sub-requirements of associated domain categories, then obtain the search results corresponding to each standardized sub-requirement based on the automotive electronic manual, and then aggregate them into the domain knowledge set of the current standardized question and send it to the answer acquisition module; the answer acquisition module has a built-in large language model, and the answer acquisition module outputs the answer corresponding to the user question based on the user question and the domain knowledge set; each module is programmed or configured to execute the steps of a data validation rule configuration method as described above.

[0014] This application also provides a computer-readable storage medium storing a computer program programmed or configured to perform an intelligent question-and-answer method for an automotive electronic manual as described above.

[0015] This application also provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of the intelligent question-and-answer method for an automotive electronic manual as described above.

[0016] The beneficial effects of this application are as follows: (1) The question-and-answer method of this application can efficiently and accurately output the answer to the user’s question based on the car electronic manual.

[0017] (2) The question-answering method of this application first uses vehicle terminology database and general automotive terminology database to term user questions to obtain standardized questions; then it annotates the standardized questions with related domains; and then it processes the search of standardized questions in related domains in parallel, thereby ensuring the completeness and comprehensiveness of the final output integrated answer.

[0018] (3) The question-answering method of this application rewrites standardized questions into the most efficient expression form (i.e., various standardized sub-requirements) in the knowledge base of the relevant domain category, based on the different storage formats of the content in different domains; and then uses these standardized sub-requirements to obtain the most accurate search results in their corresponding knowledge bases. This application significantly improves the retrieval efficiency and accuracy of standardized questions in various related domains, and further improves the accuracy and acquisition efficiency of domain knowledge sets.

[0019] (4) In the question answering method of this application, the final answer output by the large language model is not only based on the domain knowledge set containing accurate retrieval results of each related domain, but also based on vehicle background information. That is, on the basis of multi-source heterogeneous data fusion, combined with dynamic scene response, massive information is transformed into valuable and decision-making-guided "scenario insight" rather than a simple data listing, realizing end-to-end intelligent generation from "information" to "scenario-based answer"; the final output answer is no longer a fixed template, but is dynamically adjusted according to the real-time status of the user and the vehicle, which greatly improves the user experience and practicality, and enhances the level of intelligence of human-computer interaction in automobiles.

[0020] (5) In the process of rewriting user questions, the vehicle terminology library is more targeted and therefore has a higher priority than the general vehicle terminology library. Therefore, the S1 sub-step of this application is set as S12´. This can achieve the comprehensive terminology of colloquial terms in user questions based on the vehicle terminology library and the general vehicle terminology library, which is conducive to improving the accuracy of subsequent domain-specific annotations. Attached Figure Description

[0021] Figure 1 This is a flowchart of an intelligent question-answering method for automotive electronic manuals according to this application. Detailed Implementation

[0022] To make the technical solution of this application clearer and more explicit, the application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Solutions derived by those skilled in the art through equivalent substitution and conventional reasoning of the technical features of the technical solution of this application without creative effort all fall within the protection scope of this application.

[0023] This application provides an intelligent question-answering method for automotive electronic manuals, such as... Figure 1 As shown, it includes the following steps: S1, after rewriting user questions into standardized questions based on vehicle models, labels the domain categories associated with the current standardized questions; S2, rewrite the current standardization problem into standardization sub-requirements of related domain categories, and then collect the search results corresponding to each standardization sub-requirement based on the automotive electronic manual to form a domain knowledge set for the current standardization problem; S3 feeds the current user question and the set of domain knowledge into the large language model, and the large language model outputs the answer corresponding to the current user question.

[0024] Optionally, S3 also includes: feeding the current vehicle background information into the large language model.

[0025] S1 also includes the following sub-steps: S11 converts the acquired user questions into text format; at the same time, it retrieves the corresponding vehicle model terminology database based on the vehicle model code.

[0026] S12, based on the corresponding vehicle terminology library, standardized questions are obtained by rewriting vehicle terminology in user questions; S13, the current standardization problem is fed into the domain category labeling model, and the domain category labeling model labels the related domain categories of the current standardization problem; the domain categories include parameter query category, operation guide category, maintenance category, fault Q&A category, and timeliness information category.

[0027] Optionally, replace S12 with S12': S12´: Based on the corresponding vehicle model terminology library, after rewriting the vehicle model terminology in the user question, and then based on the general vehicle terminology library, rewriting the part of the user question that has not been rewritten with general terminology, we get the standardized question.

[0028] In S11: Users can input their questions via voice or text. For example, a user's question could be, "Turn on the automatic following function."

[0029] Both the vehicle model terminology database and the general vehicle terminology database are pre-defined, structured mapping databases set up by engineers and are updated regularly. Different vehicle models correspond to different vehicle model terminology databases. The vehicle model terminology database contains mappings from various colloquial terms to specific vehicle model terms; the general vehicle terminology database contains mappings from various colloquial terms to general vehicle model terms. In other words, within the vehicle model terminology database, several different colloquial terms may ultimately map to the same specific vehicle model term.

[0030] It should be noted that different model numbers of the same car may have different configurations. For example, the low-end model A may have "ACC adaptive cruise control"; while the high-end model A may have "NOA navigation assist".

[0031] In the process of rewriting user questions, the vehicle model terminology database takes priority over the general vehicle terminology database because it is more targeted. Therefore, in the S1 sub-step of this application, it is set as S12'. This allows for the comprehensive formalization of colloquial terms in user questions based on both the vehicle model terminology database and the general vehicle terminology database, which facilitates the improvement of the accuracy of subsequent domain-specific annotations. If the same colloquial term appears in both the vehicle model terminology database and the general vehicle terminology database, then we only rewrite it as vehicle model terminology.

[0032] During the rewriting process, if a word in a user's question is neither a technical term nor a colloquial noun existing in the terminology database, we perform fuzzy entity matching based on the semantic vector of that word by calculating cosine similarity.

[0033] In S13: the domain category labeling model is a trained LLM model; a standardization problem may be associated with multiple domain categories simultaneously.

[0034] For example, the standardized question "How do I turn on navigation-assisted driving?" may fall under both the "parameter query category" (querying whether the vehicle supports navigation-assisted driving function) and the "operation guide category" (querying how to operate and turn on navigation-assisted driving).

[0035] Optionally, replace S13 with S13´: S13', the current standardization problem and user interaction history are fed into the domain category labeling model, which labels the associated domain category of the current standardization problem; the domain categories include parameter query, operation guide, maintenance, fault Q&A, and timeliness information.

[0036] In S13': User interaction history consists of historical user questions and their corresponding answers. Some user questions are consecutive questions derived from several previous user questions. Therefore, user interaction history is also used as input to the domain category labeling model, giving the model "memory" capabilities. This allows the model to better understand the context and implicit user preferences of the current standardized question, improving the accuracy of the domain category labeling model in associating the current standardized question with a domain category.

[0037] In S2, "rewriting the current standardization problem into standardization sub-requirements of related domain categories, and then obtaining the search results corresponding to each standardization sub-requirement based on the automotive electronic manual" also includes the following: ① For the domain categories of parameter query and maintenance, the current standardized problem is rewritten into a first standardized sub-requirement that conforms to the structured query statement of SQL syntax; then, the first standardized sub-requirement is used to obtain the corresponding first search result based on the relational database containing parameter query and maintenance information in the automotive electronic manual.

[0038] In automotive electronic manuals, parameter query and maintenance information are generally stored in the form of relational databases. Therefore, in this embodiment, we use the relational database-aware NL2SQL rewriting method to rewrite the current standardized problem into a structured query statement that conforms to SQL syntax.

[0039] It should be noted that directly using keyword matching to obtain corresponding search results for precise numerical values ​​(such as battery life, size, and maintenance cycle) stored in a relational database is extremely difficult. Therefore, in this embodiment, the relational database is loaded as prior knowledge, and entity constraints, numerical comparison logic, and aggregation functions are extracted using a large model. This transforms the standardization problem of unstructured natural language into a structured query statement that conforms to SQL syntax (i.e., the first standardization sub-requirement).

[0040] The following example illustrates this: The standardized question is: "What is the maximum distance my car can travel on the highway?"

[0041] Using keyword matching directly for rewriting may result in many irrelevant articles being found, and the rewritten result will be "maximum range on high speed".

[0042] The rewriting method in ① of this application is as follows: The system identifies the field condition='highway' corresponding to "running on highways" and the aggregate function MAX() corresponding to "maximum"; it generates the rewritten statement: SELECT MAX(range_km) FROM vehicle_specsWHERE condition = 'highway' AND model_id = 'current_user_car'. This allows for the direct return of a unique and accurate value from the relational database with zero error.

[0043] ② For the domain category of fault question and answer, the current standardized question is rewritten into a second standardized sub-requirement with a <entity-relationship-attribute value> triple probe structure; then, the second standardized sub-requirement is used to obtain the corresponding second search results based on the knowledge graph containing fault question and answer information in the automotive electronic manual.

[0044] Troubleshooting questions in automotive electronic manuals are generally presented in the form of knowledge graphs. Therefore, in this embodiment, we rewrite the current standardized question into a graph traversal path query statement with a <entity-relationship-attribute value> triple probe structure based on the entity-relationship mapping rewriting method of knowledge graph paths.

[0045] It should be noted that the knowledge graph for fault-related Q&A content contains "fault phenomena" as well as the "causes" and "solutions" behind them in the form of node content, and the edges between nodes indicate that there is a relationship between the two nodes.

[0046] The following example illustrates this: The standardized question is: "How do I turn off the tire pressure light when it's on?"

[0047] The rewriting method in ② of this application involves: identifying the core entity "tire pressure warning light," predicting the user's intent to include "cause analysis" and "solution," and generating the rewritten statement: MATCH (n:Fault {name: 'tire pressure warning'})-[:CAUSED_BY]->(reason), (n)-[:SOLVED_BY]->(solution) RETURN reason, solution. Ultimately, by leveraging the hop count reasoning capability of the knowledge graph, not only is the method for elimination retrieved, but also potential causes of the fault are identified, achieving a deep-level response.

[0048] ③ For the category of operation guides, after removing the noise of modal words and function words in the current standardization problem, rewrite it into an instructional phrase and then vectorize it to obtain the third standardization sub-requirement; then use the third standardization sub-requirement to obtain the corresponding third search results based on the text containing operation guide information in the car electronic manual.

[0049] In this embodiment, vectorization is performed as a vector augmentation rewrite based on semantic density.

[0050] It's important to note that automotive electronic manuals often contain user guides in PDF or HTM text format. Standardization problems, however, may contain a lot of noise (meaningless interjections and function words), such as phrases like "Excuse me, could you tell me about that..." and be in interrogative sentences, while user guide text is typically declarative. If we directly vectorize the standardization problems and calculate the matching degree between the standardization problem vector and a specific text content vector using vector space distance, two vectors that should be very close (i.e., have a high matching degree) will be found to be too far apart.

[0051] Therefore, this application uses a large model to rewrite the "interrogative sentences" in the standardized questions into "standard declarative sentences" or "instructional phrases" that are consistent with the style of the titles or paragraphs in the operation guide content, and removes meaningless function words and modal words to reduce interference in the vector matching process.

[0052] The following example illustrates this: The standardization question is: "Under what circumstances can NOA be used?" The rewriting method in ③ of this application involves: removing meaningless interjections and noise from function words, completing the subject, and mimicking the style of a manual's table of contents; generating the rewritten statement: "Activation conditions and applicable scenarios for the NOA (Navigation Assistance) function". The vector obtained after rewriting highly overlaps geometrically with the vector in the "Function Introduction" section of the automotive electronic manual, significantly improving the Top-1 recall accuracy.

[0053] ④ For time-sensitive information categories, extract entity and time-sensitive keywords from the current standardization problem, then expand the synonyms of entity and time-sensitive, remove stop words, construct Boolean query logic to obtain the fourth standardization sub-requirement; then use the fourth standardization sub-requirement to obtain the corresponding fourth search results based on the search engine.

[0054] Timely information content often exceeds the scope of the car's electronic manual, requiring the use of external search APIs. However, due to the current standardization issues, the search results can become disorganized because the input consists of long and complex sentences.

[0055] The following example illustrates this: The standardized question is: "Has the price of 95-octane gasoline increased?" The rewriting method in ④ of this application is as follows: extract the entity and time-sensitive keywords in the current standardization problem: "2025 95-octane gasoline price trend today's oil price"; through keyword combination, filter out outdated historical information and guide the search engine to return the latest time-sensitive data as the corresponding fourth search result.

[0056] In S2, for the same standardization problem, the processes of rewriting the above four types of standardized sub-requirements into related domain categories and obtaining corresponding search results are parallel. These parallel search results are eventually aggregated into a set of domain knowledge for the current standardization problem.

[0057] The content in different domain categories exists in different formats (i.e., different knowledge bases). Therefore, we rewrite the same standardized problem into the most efficient expression form in the knowledge base of its associated domain category (i.e., various standardized sub-requirements); and then use these standardized sub-requirements to obtain the most accurate search results in their corresponding knowledge bases.

[0058] Obtaining "vehicle background information" also includes the following: Step 1: Obtain user profiles, real-time multi-modal user information, vehicle status information, and vehicle driving environment information.

[0059] Step 2: Use the first LLM model to output personalized user tags based on user profiles and real-time multimodal user information; at the same time, use the second LLM model to output vehicle scene tags based on vehicle status information and vehicle driving environment information.

[0060] Step 3: Then, use the third LLM model to merge the user's personalized tags and vehicle scene tags into vehicle background information.

[0061] In step 1: the user profile includes static information such as the user's driving experience, commonly used vehicle functions, and past question preferences; real-time multi-modal information includes the user's real-time status obtained through sensors such as in-vehicle cameras and microphones, such as the user's mood, whether the user is making a phone call, or whether the user is looking at navigation; vehicle status information includes real-time vehicle data, such as current speed, gear, fuel level, mileage, and fault codes; and vehicle driving environment information includes external environmental information obtained through GPS, maps, radar, etc., such as whether the vehicle is currently on a highway, on a city road, in a parking lot, and the weather conditions.

[0062] In steps 2 and 3: the first LLM model, the second LLM model, and the third LLM model are all trained LLM models. User personalization tags are pre-defined by technical personnel, such as "new driver" or "in a good mood"; vehicle scenario tags include, for example, "800 km driven, current highway conditions." Vehicle background information includes, for example, "The user is a new driver, driving conservatively, the vehicle has driven 800 km, is currently on a highway, the destination is City F, and the user is in a good mood."

[0063] Obtaining vehicle background information enables the final output to be more tailored to the person, vehicle, time, and place.

[0064] In S3, the large language model used to output the answer to the current user's question is a multi-task, multi-constraint generator. We leverage the powerful generation capabilities of the large language model, integrate vehicle background information, and the accurate retrieval results of the user's question in different related domain categories to ultimately generate a high-quality, accurate, and comprehensive natural language answer.

[0065] As shown in Table 1, 800 test cases were constructed, and the knowledge question answers of the technical solutions of this application and the prior art solutions were scored by MOS based on four dimensions: accuracy, completeness, coherence and security.

[0066] Among them, accuracy (also known as validity) measures the precision of an answer in answering a question or responding to an instruction; completeness measures the completeness of an answer in answering a question or responding to an instruction; coherence (also known as comprehensibility) measures the readability and comprehensibility of an answer in answering a question or responding to an instruction; and safety measures the safety of an answer in answering a question or responding to an instruction, reducing safety risks to users when driving, using a vehicle, or parking.

[0067] Table 1

[0068] It is evident that the technical solution of this application is superior to the question-and-answer method of existing technical solutions in terms of accuracy, completeness, coherence, and security.

[0069] The question-and-answer method described in this application can efficiently and accurately output answers to user questions based on automotive electronic manuals.

[0070] If a user's question contains automotive terminology, such as "fault code P0171," existing technologies, relying solely on text similarity searches, can easily provide incorrect answers. Similarly, if the user's question inquires about vehicle maintenance schedules, existing technologies might provide detailed maintenance information. However, this application's question-and-answer method... 1. First, user questions are terminologically processed using vehicle model terminology and general automotive terminology databases to obtain standardized questions; then, the standardized questions are labeled with related domains; and then, the standardized questions are retrieved in parallel within the related domains, thereby ensuring the completeness and comprehensiveness of the final integrated answer output.

[0071] 2. Within each related domain, based on the different storage formats of the content in different domains, the standardized problem is rewritten into the most efficient expression form (i.e., various standardized sub-requirements) in the knowledge base of its related domain category; then, these standardized sub-requirements are used to obtain the most accurate search results in their corresponding knowledge bases. This application significantly improves the retrieval efficiency and accuracy of standardized problems in each related domain, and further enhances the accuracy and retrieval efficiency of domain knowledge sets.

[0072] In the question-answering method of this application, the final answer output by the large language model is based not only on the domain knowledge set containing accurate retrieval results from various related fields, but also on vehicle background information. That is, based on the fusion of multi-source heterogeneous data and combined with dynamic scene response, massive information is transformed into valuable and decision-guiding "scenario insights" rather than a simple data listing, realizing end-to-end intelligent generation from "information" to "scenario-based answers". The final output answer is no longer a fixed template, but is dynamically adjusted according to the real-time status of the user and the vehicle, which greatly improves the user experience and practicality, and enhances the intelligence level of automotive human-machine interaction.

[0073] This application also provides an intelligent question-and-answer system for automotive electronic manuals, including: The module includes a standardized question acquisition module, a domain category annotation module, a domain knowledge set acquisition module, and an answer acquisition module. The standardized question acquisition module is used to rewrite user questions into standardized questions based on vehicle models and then send them to the domain category labeling module; at the same time, the standardized question acquisition module also sends user questions to the answer acquisition module; The domain category annotation module is used to annotate the domain categories associated with the standardization issues and then send them to the domain knowledge set acquisition module; The domain knowledge set acquisition module is used to rewrite the current standardization problem into standardization sub-requirements of related domain categories, and then, based on the automotive electronic manual, obtain the search results corresponding to each standardization sub-requirement and aggregate them into the domain knowledge set of the current standardization problem before sending it to the answer acquisition module. The answer retrieval module has a built-in large language model. Based on the user's question and the domain knowledge set, the answer retrieval module outputs the answer corresponding to the user's question. Each module is programmed or configured to perform the steps of an intelligent question-and-answer method for automotive electronic manuals as described above.

[0074] This application also provides a computer-readable storage medium storing a computer program programmed or configured to perform an intelligent question-and-answer method for an automotive electronic manual as described in Embodiment 1.

[0075] This application also provides a computer program product, including a computer program / instructions that are executed by a processor to implement the steps of an intelligent question-and-answer method for an automotive electronic manual as described in Embodiment 1.

[0076] The technologies, shapes, and structures not described in detail in this application are all well-known technologies. It should also be noted that the above are merely preferred embodiments of this application and are not intended to limit the scope of this application. The components or steps in the embodiments of this application can be decomposed and / or recombined, and these decompositions and / or recombinations should be considered as equivalent solutions of this application and should all fall within the protection scope of this application.

Claims

1. An intelligent question-answering method for automotive electronic manuals, characterized in that, Includes the following steps: S1, after rewriting user questions into standardized questions based on vehicle models, labels the domain categories associated with the current standardized questions; S2, rewrite the current standardization problem into standardization sub-requirements of related domain categories, and then collect the search results corresponding to each standardization sub-requirement based on the automotive electronic manual to form a domain knowledge set for the current standardization problem; S3 feeds the current user question and the set of domain knowledge into the large language model, and the large language model outputs the answer corresponding to the current user question.

2. The intelligent question-answering method for automotive electronic manuals according to claim 1, characterized in that, S1 also includes the following sub-steps: S11: Convert the acquired user questions into text format; at the same time, retrieve the corresponding vehicle terminology database based on the vehicle model code; S12, based on the corresponding vehicle terminology library, standardized questions are obtained by rewriting vehicle terminology in user questions; S13, the current standardization problem is fed into the domain category labeling model, and the domain category labeling model labels the related domain categories of the current standardization problem; the domain categories include parameter query category, operation guide category, maintenance category, fault Q&A category, and timeliness information category.

3. The intelligent question-answering method for automotive electronic manuals according to claim 2, characterized in that, Replace S12 with S12´: S12´, based on the corresponding vehicle terminology library, rewrite the vehicle terminology in the user question, and then based on the general vehicle terminology library, rewrite the parts of the user question that have not been rewritten with general terminology to obtain the standardized question.

4. The intelligent question-answering method for automotive electronic manuals according to claim 2, characterized in that, Replace S13 with S13´: S13´, which feeds the current standardized problem and user interaction history into the domain category labeling model. The domain category labeling model labels the associated domain categories of the current standardized problem. The domain categories include parameter query, operation guide, maintenance, fault Q&A, and timeliness information.

5. The intelligent question-answering method for automotive electronic manuals according to claim 2, characterized in that, The current standardization problem is rewritten into standardization sub-requirements of related domain categories. Then, based on the automotive electronic manual, the search results corresponding to each standardization sub-requirement are obtained, including the following: For the domain categories of parameter query and maintenance, the current standardization problem is rewritten into a first standardization sub-requirement that conforms to the structured query statement of SQL syntax; then, the first standardization sub-requirement is used to obtain the corresponding first search result based on the relational database containing parameter query and maintenance information in the automotive electronic manual. For the domain category of fault question and answer, the current standardized question is rewritten as a second standardized sub-requirement with a <entity-relationship-attribute value> triple probe structure; then, the second standardized sub-requirement is used to obtain the corresponding second search results based on the knowledge graph containing fault question and answer information in the automotive electronic manual. For the category of operation guides, after removing the noise of modal words and function words in the current standardization problem, the text is rewritten into an instructional phrase and then vectorized to obtain the third standardization sub-requirement; then, the third standardization sub-requirement is used to obtain the corresponding third search results based on the text containing operation guide information in the car electronic manual. For time-sensitive information categories, extract entities and time-sensitive keywords from the current standardization problem, then expand the synonyms of entities and time-sensitive information, remove stop words, and construct Boolean query logic to obtain the fourth standardization sub-requirement; then use the fourth standardization sub-requirement to obtain the corresponding fourth search results based on the search engine.

6. The intelligent question-answering method for automotive electronic manuals according to claim 1, characterized in that, S3 also includes: feeding the current vehicle background information into the large language model.

7. The intelligent question-answering method for automotive electronic manuals according to claim 6, characterized in that, Obtaining vehicle background information also includes the following: Step 1: Obtain user profiles, real-time multi-modal user information, vehicle status information, and vehicle driving environment information; Step 2: Use the first LLM model based on user profile and real-time multimodal user information to output personalized user tags; at the same time, use the second LLM model based on vehicle status information and vehicle driving environment information to output vehicle scene tags. Step 3: Then, use the third LLM model to merge the user's personalized tags and vehicle scene tags into vehicle background information.

8. An intelligent question-and-answer system for automotive electronic manuals, characterized in that, include: The module includes a standardized question acquisition module, a domain category annotation module, a domain knowledge set acquisition module, and an answer acquisition module. The standardized question acquisition module is used to rewrite user questions into standardized questions based on vehicle models and then send them to the domain category labeling module; at the same time, the standardized question acquisition module also sends user questions to the answer acquisition module; The domain category annotation module is used to annotate the domain categories associated with the standardization problem and then send them to the domain knowledge set acquisition module; the domain knowledge set acquisition module is used to rewrite the current standardization problem into standardization sub-requirements of associated domain categories, and then, based on the automotive electronic manual, obtain the search results corresponding to each standardization sub-requirement and aggregate them into the domain knowledge set of the current standardization problem before sending it to the answer acquisition module; the answer acquisition module has a built-in large language model, and outputs the answer corresponding to the user's question based on the user's question and the domain knowledge set; each module is programmed or configured to execute the steps of the data validation rule configuration method as described in any one of claims 1-7.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program that is programmed or configured to perform an intelligent question-and-answer method for an automotive electronic manual as described in any one of claims 1-7.

10. A computer program product comprising a computer program / instructions, characterized in that: When the computer program / instruction is executed by the processor, it implements the steps of an intelligent question-and-answer method for automotive electronic manuals as described in any one of claims 1-7.

Images