An artificial intelligence vehicle automatic question answering method and system

Abstract

The present invention proposes an artificial intelligence automobile automatic question-answering method and system, which belongs to the fields of artificial intelligence, natural language processing and human-computer interaction; the automobile question-answering knowledge base of the knowledge graph used in the method enables the relationship between entities to be quickly mapped to the corresponding position of the knowledge graph, and can quickly generate results, which improves the response speed compared with text retrieval; when analyzing data information, a deep learning model and a general large model question-answering system are combined to generate accurate relationships between entities, and a method for in-depth analysis and determination of the relationships between entities is given to ensure that the generated relationships between entities are accurate and reliable; the relationships between the generated entities are mapped to the knowledge graph automobile question-answering knowledge base, and the results generated by the knowledge graph automobile question-answering knowledge base are fed back to the car owner; therefore, the method will be more accurate in identifying data information, and the information fed back to the car owner will be more precise.

Description

Technical Field

[0001] The present invention relates to the fields of artificial intelligence, natural language processing and human-computer interaction, and in particular to an artificial intelligence vehicle automatic question-answering method and system. Background Art

[0002] Artificial intelligence car automatic question answering usually focuses on enhancing driving experience and safety; it usually includes: natural language processing (NLP), speech recognition synthesis, machine learning and deep learning technology, etc. The comprehensive application of multiple technologies not only improves the intelligence level of the vehicle, but also greatly enhances the vehicle's interactivity and personalized service capabilities, making driving safer, more efficient and more enjoyable.

[0003] Although AI-powered car question-answering has made significant progress in improving user experience and driving safety, there are still some deficiencies in technology and practical application. The deficiencies include: inaccurate recognition of data information, including inaccurate recognition of noisy in-vehicle environments and complex image information; insufficient real-time response speed. In an environment that requires a quick response, it takes a certain amount of time for the analysis and retrieval system to perform analysis, resulting in the real-time feedback speed of the response system not being fast enough, which may affect safety decisions in emergency situations. Summary of the invention

[0004] The purpose of the present invention is to overcome the above problems existing in the prior art and to greatly improve the technical effect thereof on the basis of the original technology; for this purpose, the present invention provides an artificial intelligence vehicle automatic question-answering method, the method comprising:

[0005] Car active question and answer mode and car passive question and answer mode;

[0006] The vehicle active question-and-answer mode refers to the question-and-answer mode being initiated by the vehicle. The vehicle active question-and-answer mode includes:

[0007] (1) Establishing a knowledge graph-based automotive professional question-and-answer knowledge base, and connecting the established knowledge graph-based automotive professional question-and-answer knowledge base to a general large-scale model question-and-answer system, wherein a deep learning model is provided at the front end of the knowledge graph-based automotive professional question-and-answer knowledge base; during the operation and stop of the automobile, the automobile actively acquires data information including the conditions inside and outside the automobile and the automobile itself, wherein the data information includes: image information, voice information, and text information;

[0008] (2) Inputting the acquired data information into the question-answering system of the deep learning model and the general large model respectively, and analyzing the main entity, sub-entity, and the relationship between the main entity and the sub-entity in the data information; if the main entity, sub-entity, and the relationship between the main entity and the sub-entity obtained by the analysis of the deep learning model and the general large model question-answering system are all the same, then mapping the analysis results to the automotive professional question-answering knowledge base of the knowledge graph; the automotive professional question-answering knowledge base of the knowledge graph performs query execution and answer generation operations according to the main entity, sub-entity, and the relationship between the main entity and the sub-entity, and broadcasts the generated answer through the voice controller and the video controller;

[0009] (3) If the main entity, sub-entity, and the relationship between the main entity and the sub-entity obtained by the deep learning model and the general large model question answering system are not completely the same; then the deep learning model and the general large model question answering system are used to further analyze and determine the relationship between the entities; the method for further analyzing and determining the relationship between the entities is as follows: prediction of entity and entity relationship: using various deep learning models to analyze the data information obtained in (1), and using various deep learning models to predict multiple groups of main entities, sub-entities, and the relationship between the main entity and the sub-entity corresponding to various deep learning models; generation of new entity and entity relationship: separating and counting the multiple groups of main entities, sub-entities, and the relationship between the main entity and the sub-entity, the separation counting method is: separating the main entity, sub-entity, and the relationship between the main entity and the sub-entity in each group, and respectively establishing a main entity multiset {P i}, sub-entity multiset {S i}, and the relationship multiset between the primary entity and the secondary entity {PS i}, where P represents the main entity, S represents the secondary entity, PS represents the relationship between the main entity and the secondary entity, i represents the i-th main entity, secondary entity and their relationship, i∈[1,n], i∈Z, n represents the number of deep learning models; for the multi-set {P i}, {S i} and {PS i} are counted, and multiple sets {P i}, {S i} and {PS i} count the top three entities and relationships; by combining, establish 9 combinations of the top three main entities and the top three sub-entities, and observe whether the main and sub-entities in the 9 combinations exist in the multiset {PS i} in the main and secondary entities corresponding to the top three relationships with the same count; if they exist, extract the corresponding main and secondary entities and the corresponding main and secondary entities’ relationships to generate new entities and relationships between entities; if they do not exist, take {PS i} to count the fourth relationship and observe whether the main and secondary entities exist in the multiset {PS i} the corresponding principal and secondary entities in the fourth relationship of the count; if it still does not exist, then take the multiset {PS i}The corresponding method for the fourth place in the count relationship is {PS i} count the jth relationship until the primary and secondary entities exist in the multiset {PS i} counts the corresponding primary and secondary entities in the j-th relationship, then generates a new relationship between entities, j ≥ 5, j is an integer; verification of the new relationship between entities: the established new relationship between entities is input into the general large model question-and-answer system in the form of reverse question-and-answer, and observe whether the results given by the large model contain the corresponding data information analyzed by various deep learning models; if it does, it proves that the established new relationship between entities is valid; if not, it proves that the established new relationship between entities is invalid; after establishing a valid new relationship between entities, map the new relationship between entities to the automotive professional question-and-answer knowledge base of the knowledge graph; the automotive professional question-and-answer knowledge base of the knowledge graph performs query execution and answer generation operations based on the new primary entity, secondary entity, and the relationship between the primary entity and the secondary entity, and broadcasts the generated accurate answer through the voice controller and the video controller;

[0010] Passive question-and-answer mode of automobile means that the question-and-answer is initiated by the user, including the car owner and the user who registered to use the car; when the face recognition system is installed on the car, the user needs to use the password of the face recognition system to activate the face scanning function and scan the user's face when using the car for the first time; after the face scanning system scans the face, it will save the user's facial features to the storage area of ​​the face recognition system, and the user can use the car normally; when using the car again, the user can directly use the car normally; when the user including the user uses the car, the car will use the sensor of the active question-and-answer mode to actively collect the user's facial information. If the collected facial information is in the storage area of ​​the face recognition system, it proves that the corresponding user is the user, and the user can directly use the car normally without the user actively verifying his identity; if the collected facial information is not in the storage area of ​​the face recognition system, it proves that the corresponding user is not the user, and the car will determine the user's behavior pattern according to the method of the active question-and-answer mode of the car to determine whether the user is a thief and vandal. After determining that the user is a thief and vandal, an early warning will be issued through the car connection device carried by the owner;

[0011] After confirming that the user is the registered user of the car, the car passive question and answer mode includes:

[0012] The user inputs voice and text information through the in-vehicle information system, which imports the voice and text information input by the user into the deep learning model and the general large model question and answer system; then, the voice and text information input by the user are analyzed using steps (2) and (3) of the active question and answer mode of the car, and finally the answer required by the user is generated through the professional question and answer knowledge base of the knowledge graph, and is broadcasted through the voice controller and video controller of the car.

[0013] Furthermore, a knowledge graph is a knowledge base that expresses the relationship between entities through a graphical structure. It uses graph theory to integrate and display a large amount of attribute information about entities and the relationship between entities; the entities include primary entities and secondary entities; the establishment of a professional automotive question-and-answer knowledge base based on a knowledge graph includes: demand analysis and definition: determining the application scenarios and target users of the automotive knowledge graph, and determining the types of questions that need to be answered; data collection: collecting data sources related to application scenarios and target users from existing databases, APIS, and crawler technologies; data preprocessing: using denoising technology to remove noise and irrelevant information, and using normalization to standardize Format the data to ensure data quality and consistency of format; Knowledge extraction: Apply natural language processing techniques including entity recognition and relationship extraction to identify key entities in the text and the relationships between them, such as pedestrians and vehicles, the relationship between them is the distance between vehicles; In this process, use machine learning and deep learning models to improve the accuracy of entity and relationship recognition; Knowledge modeling: Define entity types and relationship types, design the architecture of the graph through entity types and relationship types, and decide how to use standards including RDF and OWL to represent knowledge; Knowledge fusion: Integrate knowledge extracted from different sources to resolve the ambiguity of entity recognition and ensure information Consistency and integrity; Just storage and indexing: Store the constructed knowledge graph in graph databases including Neo4j and Amazon Neptune, and index the entities and relationships in the graph to improve query efficiency; Development of question-answering system: Develop a car question-answering system through the knowledge graph stored in the graph database, that is, establish a professional car question-answering knowledge base based on the knowledge graph, and integrate the developed car question-answering system into the existing car system, including the vehicle management system and the user interaction interface; Continuous update and maintenance: Update the knowledge graph regularly to include the latest data and information to ensure the accuracy and response time of the questions and answers; The general The question-and-answer system of the model includes: connecting the established automotive professional question-and-answer knowledge base based on the knowledge graph to the question-and-answer system GTP4 of the general large model to perform question-and-answer queries on data information; the front end of the automotive professional question-and-answer knowledge base of the knowledge graph is provided with a deep learning model: the set deep learning model includes: convolutional neural network CNN, recurrent neural network RNN, long short-term memory neural network and Transformer; the deep learning model is obtained by training with historical data information of the automotive industry database; for data information types, including: image information, voice information and text information, automatically matching appropriate deep learning models for data analysis.

[0014] Furthermore, if the main entity, sub-entity, and the relationship between the main entity and the sub-entity obtained through the analysis of the deep learning model and the general big model question and answer system are all the same, it includes: the main entity and sub-entity obtained by analyzing the same data information through the deep learning model and the general big model question and answer system GTP4 are consistent, and the corresponding relationship between the consistent main entity and sub-entity is also the same; any of the main entity, sub-entity and their relationship are different, that is, they are not completely the same.

[0015] Furthermore, the prediction of the entities and entity relationships includes: the entity includes a main entity and a secondary entity, and the entity relationship refers to the relationship between the main entity and the secondary entity.

[0016] Furthermore, the newly established relationships between entities are input into the general large model question and answer system in the form of a counter-question question and answer, including: using a sentence structure including: under what circumstances will a corresponding relationship be generated between the new entities and the automobile question and answer system, to ask the general large model system a counter-question, to obtain an open result of the general large model system, and to search in the result whether the corresponding data information in (2) is included; if so, it is proved that the established relationships between the new entities and the entities are valid; if not, it is proved that the established relationships between the new entities and the entities are invalid; if the established relationships between the new entities and the entities are invalid, then the generation step of the new entity and entity relationship in (3) is used to continue to generate the new entity and entity relationship, until the generated new entity and entity relationship is verified to be valid through the verification step of the new entity and entity relationship in (3).

[0017] Furthermore, the generated accurate answer will be broadcasted through the voice controller and the video controller, including: if the generated accurate answer indicates that the car is encountering an emergency or dangerous situation, the car's central control system will control the emergency braking system and the seat belt system to take corresponding self-protection measures.

[0018] Furthermore, the installation of a face recognition system on a car includes: installing any one of face recognition software including Microsoft Azure Face API, Amazon Rekognition and OpenCV on the car as the face recognition system of the present invention to store, identify and verify the user's face.

[0019] In addition, the present invention also provides an artificial intelligence automobile automatic question-answering system, which is used to implement the above-mentioned artificial intelligence automobile automatic question-answering method, and the system includes: an automobile data collection unit, an automobile data analysis unit, an analysis result display unit and a central control system; the automobile data collection unit is connected to the automobile data analysis unit, and the automobile data collection unit includes an automobile active data collection module and an automobile passive data collection module; the automobile active data collection module includes a video sensor and an audio sensor, and the automobile active data collection module is used to perform automobile active data collection in the automobile active question-answering mode, collecting data information including the inside and outside of the car and the car's own conditions, and the collected data information includes: image information, voice information and text information, and the collected data information is transmitted to the automobile data analysis unit; the automobile passive data collection module includes a touch screen display and a voice recognition system of the vehicle information system, and the user transmits information to the automobile passive data collection module by handwriting input and voice input, and the automobile passive data collection module collects the collected data The data information is transmitted to the vehicle data analysis unit; the vehicle data analysis unit is connected to the analysis result display unit, and the vehicle data analysis unit includes a deep learning module, a general large model module and a vehicle knowledge graph question and answer module; the vehicle data analysis unit first analyzes the data information through the deep learning module and the general large model module, and ensures the generation of valid entities and relationships between entities through steps (2) and (3), and then maps the generated valid entities and relationships between entities to the vehicle knowledge graph question and answer module, generates accurate results through the query execution and answer generation steps of the vehicle knowledge graph question and answer module, and transmits the generated results to the analysis result display unit; the analysis result display unit includes an on-board voice controller and a video controller, which are used to give feedback to the owner, make driving suggestions and issue warnings; the central control system is connected to the analysis result display unit, which is used to obtain emergency and dangerous situations of the vehicle according to the analysis result display unit, and take corresponding self-protection measures through the emergency braking system and seat belt system of the central control system.

[0020] The beneficial effects of the present invention are:

[0021] The present invention provides an artificial intelligence automobile automatic question-answering method and system, which divides the automobile automatic question-answering system into an automobile active question-answering mode and an automobile passive question-answering mode; in the automobile active question-answering mode, the automobile actively obtains data information inside and outside the vehicle, and transmits the data information to the question-answering system of the deep learning model and the general large model, and obtains accurate entities and relationships between entities through in-depth analysis of the deep learning model and the general large model, and maps the obtained entities and relationships between entities to the knowledge graph automobile question-answering knowledge base, and generates results through query execution and answer generation operations of the knowledge graph automobile question-answering knowledge base, and broadcasts the generated results on the voice controller and the video controller; in the automobile passive question-answering mode, the user inputs data information into the vehicle information system through text input and voice input, and the vehicle information system transmits the data information to the question-answering system of the deep learning model and the general large model, and then executes the same analysis method as the automobile active question-answering mode to analyze the data information, obtain the analysis results, and feed back the analysis results to user; the method uses a knowledge graph of a car question-and-answer knowledge base, and the relationship between entities can be quickly mapped to the corresponding position of the knowledge graph. Results can be quickly generated through query execution and answer generation, which improves the response speed compared to text retrieval; when analyzing data information, the method combines a deep learning model and a general large model question-and-answer system to generate accurate relationships between entities and entities. When the entities generated by both parties are different, a method for in-depth analysis and determination of the relationship between entities and entities is provided; the method includes: a series of steps of predicting the relationship between entities and entities, generating new entity-entity relationships and verifying the relationship between new entities and entities. In this process, the advantages of multiple deep learning technologies and general large model technologies are integrated to ensure that the relationship between generated entities and entities is accurate and reliable; the relationship between entities and entities generated by the method of the present invention is mapped to the knowledge graph car question-and-answer knowledge base, and results are generated through the knowledge graph car question-and-answer knowledge base and fed back to the car owner; the method will be more accurate in identifying data information, and the information fed back to the car owner will be more precise. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 : A flow chart of an artificial intelligence car automatic question-answering method of the present invention.

[0023] Figure 2 : A schematic diagram of an artificial intelligence automobile automatic question-answering system of the present invention. DETAILED DESCRIPTION

[0024] The specific embodiments of the present invention are described in detail below in conjunction with the accompanying drawings; it should be understood that the specific embodiments given here are only used to illustrate and explain the present invention and cannot be used to limit the present invention.

[0025] like Figure 1As shown in the figure: is a flow chart of an artificial intelligence automobile automatic question-answering method according to the present invention, the flow chart includes: an automobile active question-answering mode and an automobile passive question-answering mode; wherein the automobile active question-answering mode includes: step S100, establishing a professional automobile question-answering knowledge base based on a knowledge graph, and connecting the established professional automobile question-answering knowledge base based on a knowledge graph to a question-answering system of a general large model, wherein a deep learning model is arranged at the front end of the professional automobile question-answering knowledge base of the knowledge graph; during the operation and stop of the automobile, the automobile actively acquires data information including the conditions inside the automobile, outside the automobile and the automobile itself, wherein the data information includes: image information, voice information and text information; step S101, inputting the acquired data information into the deep The question-answering system of the deep learning model and the general large model analyzes the main entity, the sub-entity, and the relationship between the main entity and the sub-entity in the data information; if the main entity, the sub-entity, and the relationship between the main entity and the sub-entity obtained by the analysis of the deep learning model and the general large model question-answering system are all the same, the analysis result is mapped to the automotive professional question-answering knowledge base of the knowledge graph; the automotive professional question-answering knowledge base of the knowledge graph performs query execution and answer generation operations according to the main entity, the sub-entity, and the relationship between the main entity and the sub-entity, and broadcasts the generated answer through the voice controller and the video controller; step S102, if the main entity, the sub-entity, and the relationship between the main entity and the sub-entity obtained by the analysis of the deep learning model and the general large model question-answering system and the relationship between the main entity and the secondary entity is not exactly the same; the deep learning model and the general large model question-answering system are used to further analyze and determine the relationship between the entities; the method for further analyzing and determining the relationship between the entities includes: predicting the relationship between the entities, generating new entity-entity relationships and verifying the new entity-entity relationships; establishing effective new entity-entity relationships by deeply analyzing and determining the relationship between the entities; after establishing the effective new entity-entity relationships, mapping the new entity-entity relationships to the automotive professional question-answering knowledge base of the knowledge graph; the automotive professional question-answering knowledge base of the knowledge graph is based on the new main entity, secondary entity, and main entity The relationship between the entity and the sub-entity is used to perform query execution and answer generation operations, and the generated accurate answer is broadcasted through the voice controller and the video controller; in addition, the car passive question and answer mode includes: step S200, the car passive question and answer mode refers to that the question and answer is initiated by the user, and the user includes the car owner and the user who registers to use the car; the face recognition system is installed on the car, and when the user wants to use the car for the first time, the user needs to use the password of the face recognition system to activate the face scanning function and scan the user's face; after the face scanning system scans the face, it will save the user's facial features to the storage area of ​​the face recognition system, and the user can use the car normally; when using the car again, the user can directly use the car normally;When a user, including the user, uses the car, the car will use the sensor in the active question-and-answer mode to actively collect the user's facial information. If the collected facial information is in the storage area of ​​the facial recognition system, it proves that the corresponding user is the user, and the user can directly use the car normally without the user actively verifying his identity; if the collected facial information is not in the storage area of ​​the facial recognition system, it proves that the corresponding user is not the user, and the car will determine the user's behavior pattern based on the method of the car's active question-and-answer mode to determine whether the user is a thief and vandal. After determining that the user is a thief and vandal, the owner will carry it with him. The vehicle-connected device issues an early warning; Step S201, after confirming that the user is a registered user of the car, the user inputs voice information and text information through the vehicle information system, and the vehicle information system imports the voice information and text information input by the user into the deep learning model and the general large model question and answer system; Subsequently, the voice information and text information input by the user are analyzed using the steps S101 and S102 of the vehicle active question and answer mode, and finally the answers required by the user are generated through the professional question and answer knowledge base of the knowledge graph, and are broadcasted through the voice controller and video controller on the vehicle. ;

[0026] Specifically, the automobile automatic question-and-answer system is divided into an automobile active question-and-answer mode and an automobile passive question-and-answer mode; in the automobile active question-and-answer mode, the automobile actively obtains data information inside and outside the car, and transmits the data information to the question-and-answer system of the deep learning model and the general large model, and obtains accurate entities and relationships between entities through in-depth analysis of the deep learning model and the general large model, and maps the acquired entities and relationships between entities to the knowledge graph automobile question-and-answer knowledge base, and generates results through query execution and answer generation operations of the knowledge graph automobile question-and-answer knowledge base, and broadcasts the generated results on the voice controller and the video controller; in the automobile passive question-and-answer mode, the user inputs data information into the vehicle information system through text input and voice input, and the vehicle information system transmits the data information to the question-and-answer system of the deep learning model and the general large model, and then executes the same analysis method as the automobile active question-and-answer mode to analyze the data information, obtain analysis results, and feed back the analysis results to the user.

[0027] Car active question and answer mode:

[0028] Step S100, establish a vehicle-use professional question and answer knowledge base based on the knowledge graph, and connect the established vehicle-use professional question and answer knowledge base based on the knowledge graph to the question and answer system of the general large model, and the front end of the vehicle-use professional question and answer knowledge base of the knowledge graph is provided with a deep learning model; during the operation and stop of the car, the car actively obtains data information including the conditions inside and outside the car and the car itself, and the data information includes: image information, voice information and text information.

[0029] In the above embodiment, specifically, the knowledge graph is a knowledge base that expresses the relationship between entities through a graphical structure. It uses graph theory to integrate and display a large amount of attribute information about entities and the relationship between entities; the entities include main entities and secondary entities; the establishment of a professional automotive question-and-answer knowledge base based on the knowledge graph includes: demand analysis and definition: determining the application scenarios and target users of the automotive knowledge graph, and determining the types of questions that need to be answered; data collection: collecting data sources related to the application scenarios and target users from existing databases, APIS and crawler technologies; data preprocessing: using denoising technology to remove noise and irrelevant information, and using normalization to standardize the data format to ensure data quality and format consistency. Consistency; Knowledge extraction: Apply natural language processing technology including entity recognition and relationship extraction to identify key entities in the text and the relationship between them, such as pedestrians and vehicles, the relationship between them is the distance between vehicles; In this process, use machine learning and deep learning models to improve the accuracy of entity and relationship recognition; Knowledge modeling: Define entity types and relationship types, design the architecture of the graph through entity types and relationship types, and decide how to use standards including RDF and OWL to represent knowledge; Knowledge fusion: Integrate knowledge extracted from different sources to solve the ambiguity of entity recognition and ensure the consistency and integrity of information; Just store and index: Store the constructed knowledge graph in Neo4j and Amazon Neptune's graph database, and index the entities and relationships in the graph to improve query efficiency; development of the question-and-answer system: develop a vehicle question-and-answer system through the knowledge graph stored in the graph database, that is, establish a vehicle-use professional question-and-answer knowledge base based on the knowledge graph, and integrate the developed vehicle-use question-and-answer system into the existing system of the car, including the vehicle management system and the user interaction interface; continuous update and maintenance: regularly update the knowledge graph to include the latest data and information to ensure the accuracy and response time of the questions and answers; the question-and-answer system of the general large model includes: connecting the established vehicle-use professional question-and-answer knowledge base based on the knowledge graph to the question-and-answer system GTP4 of the general large model to perform question-and-answer queries on data information; the front end of the vehicle-use professional question-and-answer knowledge base of the knowledge graph is provided with a deep learning model: the set deep learning models include: convolutional neural network CNN, recurrent neural network RNN, long short-term memory neural network and Transformer, etc.; the deep learning model is obtained by training the historical data information of the automotive industry database; according to the data information type, including: image information, voice information and text information, the appropriate deep learning model is automatically matched for data analysis.

[0030] Step S101, input the acquired data information into the question and answer system of the deep learning model and the general large model respectively, and analyze the main entity, sub-entity, and the relationship between the main entity and the sub-entity in the data information; if the main entity, sub-entity, and the relationship between the main entity and the sub-entity obtained by the analysis of the deep learning model and the general large model question and answer system are all the same, the analysis result is mapped to the automotive professional question and answer knowledge base of the knowledge graph; the automotive professional question and answer knowledge base of the knowledge graph performs query execution and answer generation operations according to the main entity, sub-entity, and the relationship between the main entity and the sub-entity, and broadcasts the generated answer through the voice controller and the video controller.

[0031] In the above embodiment, specifically, the main entity, sub-entity, and the relationship between the main entity and the sub-entity obtained through the analysis of the deep learning model and the general large model question and answer system are all the same, which means: the main entity and sub-entity obtained by analyzing the same data information through the deep learning model and the general large model question and answer system GTP4 are consistent, and the corresponding relationship between the consistent main entity and sub-entity is also the same; any of the main entity, sub-entity and their relationship are different, that is, they are not completely the same; for example: the main entity car and the sub-entity pedestrian in front of the sub-entity obtained through the deep learning model and the general large model analysis are all the same, and the distance between the main entity car and the pedestrian in front of the sub-entity is approximately 20 meters (with a tolerance of no more than 10 centimeters), then it is considered that the main entity, sub-entity, and the relationship between the main entity and the sub-entity obtained through the analysis of the deep learning model and the general large model question and answer system are all the same.

[0032] Step S102, if the main entity, sub-entity, and the relationship between the main entity and the sub-entity obtained by the analysis of the deep learning model and the general large model question and answer system are not exactly the same; then the deep learning model and the general large model question and answer system are used to continue to deeply analyze and determine the relationship between the entities; the method for continuing to deeply analyze and determine the relationship between the entities includes: prediction of the relationship between the entities, generation of new entity and entity relationships, and verification of the new entity and entity relationships; establishing effective new entity and entity relationships through in-depth analysis and determination of the relationship between the entities; after establishing effective new entity and entity relationships, mapping the new entity and entity relationships to the automotive professional question and answer knowledge base of the knowledge graph; the automotive professional question and answer knowledge base of the knowledge graph performs query execution and answer generation operations according to the new main entity, sub-entity, and the relationship between the main entity and the sub-entity, and broadcasts the generated accurate answer through the voice controller and the video controller.

[0033] In the above embodiment, specifically, the prediction of entities and entity relationships includes: using various deep learning models to analyze the data information obtained in step S100, and predicting the results of multiple groups of main entities, sub-entities, and relationships between main entities and sub-entities corresponding to various deep learning models through various deep learning models; the generation of new entities and entity relationships includes: separating and counting the obtained multiple groups of main entities, sub-entities, and relationships between main entities and sub-entities, and the separation counting method is: separating the main entities, sub-entities, and relationships between main entities and sub-entities in each group, and respectively establishing main entity multisets {P i}, sub-entity multiset {S i}, and the relationship multiset between the primary entity and the secondary entity {PS i}, where P represents the main entity, S represents the secondary entity, PS represents the relationship between the main entity and the secondary entity, i represents the i-th main entity, secondary entity and their relationship, i∈[1,n], i∈Z, n represents the number of deep learning models; for the multi-set {P i}, {S i} and {PS i} are counted, and multiple sets {P i}, {S i} and {PS i} count the top three entities and relationships; by combining, establish 9 combinations of the top three main entities and the top three sub-entities, and observe whether the main and sub-entities in the 9 combinations exist in the multiset {PS i} in the main and secondary entities corresponding to the top three relationships with the same count; if they exist, extract the corresponding main and secondary entities and the corresponding main and secondary entities’ relationships to generate new entities and relationships between entities; if they do not exist, take {PS i} to count the fourth relationship and observe whether the main and secondary entities exist in the multiset {PS i} the corresponding principal and secondary entities in the fourth relationship of the count; if it still does not exist, then take the multiset {PS i}The corresponding method for the fourth place in the count relationship is {PS i} count the jth relationship until the primary and secondary entities exist in the multiset {PS i}, a new relationship between entities is generated by counting the corresponding primary and secondary entities in the j-th relationship of the count, j≥5, j is an integer; the verification of the new relationship between entities includes: inputting the established new relationship between entities into the question-answering system of the general large model in the form of reverse question-answering, and observing whether the results given by the large model contain the corresponding data information analyzed by various deep learning models; if contained, it proves that the established new relationship between entities is valid; if not, it proves that the established new relationship between entities is invalid.

[0034] In the above embodiment, specifically, nine combinations of the top three main entities and the top three sub-entities are established, for example, the top three main entities are P1, P2 and P3, and the top three sub-entities are S1, S2 and S3, then the nine combinations are P1S1, P1S2, P1S3, P2S1, P2S2, P2S3, P3S1, P3S2 and P3S3; the observation of whether the main and sub-entities in the nine combinations exist in the multiset {PS i} The top three relationships correspond to the main and secondary entities, and by observing the multiset {PS i}, whether the principal and subsidiary entities corresponding to the top three relationships in terms of count correspond to the 9 principal and subsidiary entities given above; if the multiset {PS i}, the top three relationships in terms of count correspond to a set of primary and secondary entities P2 and S2 respectively. The combination P2S2 is exactly the same as {PS i}, a relationship between the top three entities in terms of count corresponds to the primary entity P2 and the secondary entity S2, and a new relationship between entities is generated.

[0035] In the above embodiment, specifically, the prediction of the entity and entity relationship includes: the entity includes a main entity and a secondary entity, and the entity relationship refers to the relationship between the main entity and the secondary entity.

[0036] In the above embodiment, specifically, the established new relationship between entities is input into the question-answering system of the general large model in the form of a counter-question question and answer method: the sentence pattern including: under what circumstances will the new entity and the entity have a corresponding relationship, the general large model system is asked to obtain an open result of the general large model system, and the result is searched to see whether it contains the corresponding data information in step S101. If it does, it proves that the established new relationship between entities is valid; if it does not, it proves that the established new relationship between entities is invalid; if the established new relationship between entities is invalid, the generation step of the new entity and entity relationship in step S102 is adopted to continue to gradually obtain the multi-set {PS i}, continue to generate new entities and relationships between entities until the generated new relationships between entities are verified to be valid through the verification step of the new relationships between entities in step S102.

[0037] In the above embodiment, specifically, the generated accurate answer will be broadcasted through the voice controller and the video controller, including: if the generated accurate answer indicates that the car is encountering an emergency or dangerous situation, the car's central control system will control the emergency braking system and the seat belt system to take corresponding self-protection measures.

[0038] Passive Q&A mode for cars:

[0039] Step S200, the passive question-and-answer mode of the car means that the question-and-answer is initiated by the user, and the user includes the car owner and the user who registers to use the car; the face recognition system is installed on the car, and when the user wants to use the car for the first time, the user needs to use the password of the face recognition system to activate the face scanning function and scan the user's face; after the face scanning system scans the face, it will save the user's facial features to the storage area of ​​the face recognition system, and the user can use the car normally; when using the car again, the user can directly use the car normally; when the user including the user uses the car, the car will use the sensor of the active question-and-answer mode to actively collect the user's facial information. If the collected facial information is in the storage area of ​​the face recognition system, it proves that the corresponding user is the user, and the user can directly use the car normally without the user actively verifying his identity; if the collected facial information is not in the storage area of ​​the face recognition system, it proves that the corresponding user is not the user, and the car will determine the user's behavior pattern according to the method of the active question-and-answer mode of the car to determine whether the user is a thief and a vandal. After determining that the user is a thief and a vandal, an early warning is issued through the car connection device carried by the car owner.

[0040] In the above embodiment, specifically, the installation of a face recognition system on a car includes: installing any one of the face recognition software including Microsoft Azure Face API, Amazon Rekognition and OpenCV on the car as the face recognition system of the present invention to store, identify and verify the user's face.

[0041] Step S201, after determining that the user is a registered user of the car, the user inputs voice information and text information through the vehicle information system, and the vehicle information system imports the voice information and text information input by the user into the deep learning model and the general large model question and answer system; then, the voice information and text information input by the user are analyzed using steps S101 and S102 of the car active question and answer mode, and finally the answers required by the user are generated through the professional automotive question and answer knowledge base of the knowledge graph, and broadcast through the vehicle's voice controller and video controller.

[0042] In the above embodiment, the in-vehicle information system will import the voice information and text information input by the user into the question and answer system of the deep learning model and the general large model. After analysis by the deep learning model and the general large model, if the results obtained by both parties are completely consistent, the main entity, the sub-entity and the relationship between the main and sub-entities are directly mapped to the vehicle question and answer knowledge base of the knowledge graph, and the vehicle question and answer knowledge base directly generates the result, and the result is fed back to the user through the voice controller and the video controller; if the results obtained by both parties are not completely consistent, the method of generating new relationships between entities in step S102 is used to generate new relationships between entities; and the generated new relationships between entities are mapped to the vehicle question and answer knowledge base of the knowledge graph, and the vehicle question and answer knowledge base of the knowledge graph gives the result, and the result is fed back to the user.

[0043] like Figure 2As shown: is a schematic diagram of an artificial intelligence automobile automatic question-answering system of the present invention; the schematic diagram includes: an automobile data collection unit S30, an automobile data analysis unit S40, an analysis result display unit S50 and a central control system S60; wherein the automobile data collection unit S30 is connected to the automobile data analysis unit S40, and the automobile data collection unit S30 includes an automobile active data collection module S301 and an automobile passive data collection module S302; the automobile active data collection module S301 includes a video sensor and an audio sensor, and the automobile active data collection module S301 is used for automobile active data collection in the automobile active question-answering mode, collecting data information including the inside and outside of the car and the car's own conditions, and the collected data information includes: image information, voice information and text information, and the collected data information is transmitted to the automobile data analysis unit S40; the automobile passive data collection module S302 includes a touch screen display and a voice recognition system of the vehicle information system, and the user transmits information to the automobile passive data collection module S302 by handwriting input and voice input, and the automobile passive data collection module S302 transmits the collected data information to the automobile data analysis unit S40 The automobile data analysis unit S40 is connected to the analysis result display unit S50, and the automobile data analysis unit S40 includes a deep learning module S401, a general large model module S402 and a vehicle knowledge graph question and answer module S403; the automobile data analysis unit S40 first analyzes the data information through the deep learning module S401 and the general large model module S402, and ensures the generation of valid entities and relationships between entities through steps S101 and S102, and then maps the generated valid entities and relationships between entities to the vehicle knowledge graph question and answer module S403, through Through the query execution and answer generation steps of the vehicle knowledge graph question and answer module S403, accurate results are generated and transmitted to the analysis result display unit S50; the analysis result display unit S50 includes an on-board voice controller and a video controller, which are used to give feedback to the car owner, make driving suggestions and issue warnings; the central control system S60 is connected to the analysis result display unit S50, and is used to obtain emergency and dangerous situations of the car according to the analysis result display unit S50, and take corresponding self-protection measures through the emergency braking system and seat belt system of the central control system S60.

[0044] It should be understood that the above-mentioned embodiments are one or more embodiments of the present invention, and there are many other embodiments and variations thereof based on the present invention; the variations and modifications made by ordinary technicians in this industry through the present invention without making pioneering innovations all fall within the scope of protection of the present invention.

Claims

1. An artificial intelligence vehicle automatic question answering method, characterized in that: The method comprises: Car active question and answer mode and car passive question and answer mode; The vehicle active question-and-answer mode refers to the question-and-answer mode being initiated by the vehicle. The vehicle active question-and-answer mode includes: 1) Establishing a knowledge graph-based automotive professional question-and-answer knowledge base, and connecting the established knowledge graph-based automotive professional question-and-answer knowledge base to a general large-scale model question-and-answer system, wherein a deep learning model is provided at the front end of the knowledge graph-based automotive professional question-and-answer knowledge base; during the operation and stop of the automobile, the automobile actively acquires data information including the conditions inside and outside the automobile and the automobile itself, wherein the data information includes: image information, voice information and text information; 2) Input the acquired data information into the question-answering system of the deep learning model and the general large model respectively, and analyze the main entity, sub-entity, and the relationship between the main entity and the sub-entity in the data information; if the main entity, sub-entity, and the relationship between the main entity and the sub-entity obtained by the analysis of the deep learning model and the general large model question-answering system are all the same, then map the analysis results to the automotive professional question-answering knowledge base of the knowledge graph; the automotive professional question-answering knowledge base of the knowledge graph performs query execution and answer generation operations according to the main entity, sub-entity, and the relationship between the main entity and the sub-entity, and broadcasts the generated answer through the voice controller and the video controller; 3) If the main entity, sub-entity, and the relationship between the main entity and the sub-entity obtained by the deep learning model and the general large model question answering system analysis are not completely the same; then the deep learning model and the general large model question answering system are used to continue to deeply analyze and determine the relationship between the entities; the method for continuing to deeply analyze and determine the relationship between the entities is: prediction of entity and entity relationship: using various deep learning models to analyze the data information obtained in 1), and predicting multiple groups of main entities, sub-entities, and the relationship between the main entity and the sub-entity corresponding to various deep learning models through various deep learning models; generation of new entity and entity relationship: separate and count the obtained multiple groups of main entities, sub-entities, and the relationship between the main entity and the sub-entity, and the method of separate counting is: separate the main entity, sub-entity, and the relationship between the main entity and the sub-entity in each group, and respectively establish a main entity multiset {P i }, sub-entity multiset {S i }, and the relationship multiset between the primary entity and the secondary entity {PS i }, where P represents the main entity, S represents the secondary entity, PS represents the relationship between the main entity and the secondary entity, i represents the i-th main entity, secondary entity and their relationship, i∈[1,n], i∈Z, n represents the number of deep learning models; for the multiset {P i }、{S i } and}PS i }, and extract multiple sets {P i }、{S i } and}PS i } count the top three entities and relationships; by combining, establish 9 combinations of the top three main entities and the top three sub-entities, and observe whether the main and sub-entities in the 9 combinations exist in the multiset {PS i } in the main and secondary entities corresponding to the top three relationships with the same count; if they exist, extract the corresponding main and secondary entities and the corresponding main and secondary entities’ relationships to generate new entities and relationships between entities; if they do not exist, take {PS i } to count the fourth relationship and observe whether the main and secondary entities exist in the multiset {PS i }The corresponding principal and secondary entities in the fourth-ranked relationship of}; if it still does not exist, take the multiset {PS i The method corresponding to the fourth-ranked relationship of} is {PS i } until the primary and secondary entities exist in the multiset {PS i}, the corresponding main and secondary entities in the j-th relationship of the count of} are generated, and a new relationship between entities is generated, j≥5, j is an integer; verification of the relationship between new entities: the established relationship between new entities and entities is input into the question-and-answer system of the general large model in the form of reverse question-and-answer, and it is observed whether the results given by the large model contain the corresponding data information analyzed by various deep learning models; if it does, it proves that the established relationship between new entities and entities is valid; if not, it proves that the established relationship between new entities and entities is invalid; after establishing a valid relationship between new entities and entities, the relationship between new entities and entities is mapped to the automotive professional question-and-answer knowledge base of the knowledge graph; the automotive professional question-and-answer knowledge base of the knowledge graph performs query execution and answer generation operations based on the new main entity, secondary entity, and the relationship between the main entity and secondary entity, and broadcasts the generated accurate answer through the voice controller and the video controller; The passive question-and-answer mode of the car means that the question-and-answer is initiated by the user, including the car owner and the user who registers to use the car; the face recognition system is installed on the car, and when the user wants to use the car for the first time, the user needs to use the password of the face recognition system to activate the face scanning function and scan the user's face; after the face scanning system scans the face, it will save the user's facial features to the storage area of ​​the face recognition system, and the user can use the car normally; when using the car again, the user can directly use the car normally; when the user including the user uses the car, the car will use the sensor of the active question-and-answer mode to actively collect the user's facial information. If the collected facial information is in the storage area of ​​the face recognition system, it proves that the corresponding user is the user, and the user can directly use the car normally without the user actively verifying his identity; if the collected facial information is not in the storage area of ​​the face recognition system, it proves that the corresponding user is not the user, and the car will determine the user's behavior pattern according to the method of the active question-and-answer mode of the car to determine whether the user is a thief and a vandal. After determining that the user is a thief and a vandal, an early warning is issued through the car connection device carried by the car owner; After confirming that the user is the registered user of the car, the car passive question and answer mode includes: The user inputs voice and text information through the in-vehicle information system, which imports the voice and text information input by the user into the deep learning model and the general large model question and answer system; then, the voice and text information input by the user are analyzed using steps 2) and 3) of the car's active question and answer mode, and finally the answers required by the user are generated through the car's professional question and answer knowledge base of the knowledge graph, and broadcast through the in-vehicle voice controller and video controller.

2. The method for automatic question answering of an artificial intelligence vehicle according to claim 1, characterized in that: The knowledge graph includes: the knowledge graph is a knowledge base that expresses the relationship between entities through a graphical structure. It uses graph theory to integrate and display a large amount of attribute information about entities and the relationship between entities; the entities include main entities and secondary entities; the establishment of a professional automotive question-and-answer knowledge base based on the knowledge graph includes: demand analysis and definition: determining the application scenarios and target users of the automotive knowledge graph, and determining the types of questions that need to be answered; data collection: collecting data sources related to the application scenarios and target users from existing databases, APIS and crawler technologies; data preprocessing: using denoising technology to remove noise and irrelevant information, and using normalization to standardize the data format to ensure consistency in data quality and format Knowledge extraction: Apply natural language processing techniques including entity recognition and relationship extraction to identify key entities in the text and the relationships between them, such as pedestrians and vehicles, and the relationship between them is the distance between vehicles; in this process, use machine learning and deep learning models to improve the accuracy of entity and relationship recognition; Knowledge modeling: Define entity types and relationship types, design the architecture of the graph through entity types and relationship types, and decide how to use standards including RDF and OWL to represent knowledge; Knowledge fusion: Integrate knowledge extracted from different sources to solve the ambiguity of entity recognition and ensure the consistency and integrity of information; Knowledge storage and indexing: Store the constructed knowledge graph in Neo4j and Amazon Neptune's graph database, and index the entities and relationships in the graph to improve query efficiency; development of the question-and-answer system: develop a vehicle question-and-answer system through the knowledge graph stored in the graph database, that is, establish a vehicle-use professional question-and-answer knowledge base based on the knowledge graph, and integrate the developed vehicle-use question-and-answer system into the existing system of the car, including the vehicle management system and the user interaction interface; continuous update and maintenance: regularly update the knowledge graph to include the latest data and information to ensure the accuracy and response time of the questions and answers; the question-and-answer system of the general large model includes: connecting the established vehicle-use professional question-and-answer knowledge base based on the knowledge graph to the question-and-answer system GTP4 of the general large model to perform question-and-answer queries on data information; the front end of the vehicle-use professional question-and-answer knowledge base of the knowledge graph is provided with a deep learning model: the set deep learning models include: convolutional neural network CNN, recurrent neural network RNN, long short-term memory neural network and Transformer; the deep learning model is obtained by training the historical data information of the automotive industry database; according to the data information type, including: image information, voice information and text information, the appropriate deep learning model is automatically matched for data analysis.

3. The method for automatic question answering of an artificial intelligence vehicle according to claim 1, characterized in that: If the main entity, sub-entity, and the relationship between the main entity and the sub-entity obtained through the analysis of the deep learning model and the general big model question and answer system are all the same, it includes: the main entity and sub-entity obtained by analyzing the same data information through the deep learning model and the general big model question and answer system GTP4 are consistent, and the corresponding relationship between the consistent main entity and sub-entity is also the same; any of the main entity, sub-entity and their relationship are different, that is, they are not completely the same.

4. The method for automatic question answering of an artificial intelligence vehicle according to claim 1, characterized in that: The prediction of entities and entity relationships includes: entities include main entities and secondary entities, and entity relationships refer to the relationships between the main entities and the secondary entities.

5. The method for automatic question answering of an artificial intelligence vehicle according to claim 1, characterized in that: The newly established relationships between entities are input into the general large model question and answer system in the form of counter-question question and answer, including: using a sentence pattern including: under what circumstances will a corresponding relationship be generated between the new entities and the automobile question and answer system to ask the general large model system, to obtain an open result of the general large model system, and to search in the result whether the corresponding data information in 2) is included, if included, it proves that the established relationship between the new entities and the entities is valid; if not, it proves that the established relationship between the new entities and the entities is invalid; if the established relationship between the new entities and the entities is invalid, then the generation step of the new entity and entity relationship in 3) is used to continue to generate the new entity and entity relationship, until the generated new entity and entity relationship is verified to be valid through the verification step of the new entity and entity relationship in 3).

6. The method for automatic question answering of an artificial intelligence vehicle according to claim 1, characterized in that: The generated accurate answer will be broadcasted through the voice controller and the video controller, including: if the generated accurate answer indicates that the car is in an emergency or dangerous situation, the car central control system will control the emergency braking system and the seat belt system to take corresponding self-protection measures.

7. The method for automatic question answering of an artificial intelligence vehicle according to claim 1, characterized in that: The step of installing a face recognition system on a car includes: installing any one of face recognition software including Microsoft Azure Face API, Amazon Rekognition and OpenCV on the car as the face recognition system to store, identify and verify the user's face.

8. An artificial intelligence vehicle automatic question-answering system, the system being used to implement an artificial intelligence vehicle automatic question-answering method in claim 1, the system comprising: Automobile data collection unit, automobile data analysis unit, analysis result display unit and central control system; The vehicle data collection unit is connected to the vehicle data analysis unit, and the vehicle data collection unit includes a vehicle active data collection module and a vehicle passive data collection module; The vehicle active data collection module includes a video sensor and an audio sensor. The vehicle active data collection module is used to collect active data of the vehicle in the vehicle active question-answering mode, collect data information including the inside and outside of the vehicle and the vehicle's own conditions, and the collected data information includes: image information, voice information and text information, and transmit the collected data information to the vehicle data analysis unit; The automobile passive data collection module includes a touch screen display and a voice recognition system of the vehicle information system. The user transmits information to the automobile passive data collection module by handwriting input and voice input, and the automobile passive data collection module transmits the collected data information to the automobile data analysis unit; the automobile data analysis unit is connected to the analysis result display unit, and the automobile data analysis unit includes a deep learning module, a general large model module and a vehicle knowledge graph question and answer module; the automobile data analysis unit first analyzes the data information through the deep learning module and the general large model module, and ensures the generation of valid entities and relationships between entities through steps 2) and 3), and then maps the generated valid entities and relationships between entities to the vehicle knowledge graph question and answer module, generates accurate results through the query execution and answer generation steps of the vehicle knowledge graph question and answer module, and transmits the generated results to the analysis result display unit; the analysis result display unit includes an on-board voice controller and a video controller, which are used to give feedback to the owner, make driving suggestions and issue warnings; the central control system is connected to the analysis result display unit, and is used to obtain emergency and dangerous situations of the automobile according to the analysis result display unit, and take corresponding self-protection measures through the emergency braking system and seat belt system of the central control system.

Images