Automobile fault knowledge graph construction method and system
By generalizing entities from the automotive fault knowledge graph and constructing an entity semantic index model, the problems of high matching difficulty and low efficiency in existing technologies are solved, achieving efficient and accurate entity name matching and knowledge graph updates.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CHONGQING CHANGAN AUTOMOBILE CO LTD
- Filing Date
- 2023-11-01
- Publication Date
- 2026-07-31
AI Technical Summary
In existing technologies, matching automotive fault knowledge graphs is difficult, computationally complex, and inefficient, and cannot effectively solve the problem of matching synonyms and near-synonyms in entity names.
By generalizing the fault entities in the initial automotive fault knowledge graph, generating generalized names using a word vector model, and constructing an entity semantic index model, the word vectors of newly added entities are calculated using the entity semantic index model, simplifying the matching process, reducing matching difficulty, and improving efficiency.
It enables synonym and near-synonym matching for entity names, simplifies the matching process, improves the generalization ability and accuracy of knowledge graphs, simplifies similarity calculation, and improves efficiency and accuracy.
Smart Images
Figure CN117474093B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of knowledge graph technology, specifically to a method and system for constructing an automotive fault knowledge graph. Background Technology
[0002] With the gradual maturation of knowledge graph technology, it has been widely used in industries such as finance, e-commerce, and education. In recent years, the field of intelligent vehicles has also been exploring the use of knowledge graph technology in scenarios such as automobile manufacturing, automobile repair, and automobile marketing.
[0003] Currently, most technologies only describe the application of automotive fault knowledge graphs, without detailing how to construct them. For example, patent CN114691831A proposes a task-oriented intelligent question-and-answer system for automotive faults based on knowledge graphs, which autonomously answers user questions based on the content of the knowledge graph. Patent CN115033679A proposes a method for searching automotive repair data based on knowledge graphs, which identifies fault pairs from user input information, inputs the fault pairs and their occurrence conditions into the knowledge graph to generate candidate answers, further sorts the candidate answers, and finally outputs the N most matching answers.
[0004] Existing technology discloses a search retrieval method and system based on knowledge graph representation learning. This method uses knowledge graph representation learning to obtain vectors corresponding to entities in the graph, builds an index library corresponding to target documents based on these vectors, and then achieves the process of obtaining search results from user input of a query statement through entity linking, vector querying and retrieval, fine-grained vector similarity calculation, and document retrieval. This improves the semantic retrieval capability of search retrieval and achieves cold-start effect. Representation learning can obtain vectors corresponding to entities and accurately match entities in the query statement through entity linking. Compared with existing technologies, it can solve the problem of inaccurate retrieval due to word segmentation, achieving effective document retrieval in scenarios of "one word with multiple meanings" and "multiple words with one meaning," while reducing the dependence of the training model on the corpus during the retrieval process and improving interpretability while ensuring response time. However, this patent has weak knowledge graph generalization ability and high matching difficulty. Furthermore, this patent first performs entity matching and then queries the index library to obtain accurate vectors and candidate similar vectors, requiring similarity calculation, which is complex and inefficient. Summary of the Invention
[0005] The purpose of this invention is to provide a method and system for constructing an automotive fault knowledge graph, which addresses the problems of high matching difficulty and computational complexity in existing technologies by providing a method and system for constructing an automotive fault knowledge graph that reduces matching difficulty and increases efficiency.
[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:
[0007] A method for constructing an automotive fault knowledge graph includes:
[0008] S1: Obtain initial vehicle fault knowledge, extract fault entities and the relationships between multiple fault entities from the initial vehicle fault knowledge, and generate an initial vehicle fault knowledge graph.
[0009] S2: Generalize the original names of fault entities in the initial vehicle fault knowledge graph using a word vector model to obtain generalized names of fault entities. Store the generalized names and original names of fault entities in the initial vehicle fault knowledge graph and establish the relationship between the generalized names and original names of fault entities to generate a generalized vehicle fault knowledge graph.
[0010] S3: Obtain the original names of each fault entity from the generalized automotive fault knowledge graph, and form a fault entity original name dataset from the original names of each fault entity; obtain the generalized names of each fault entity from the generalized automotive fault knowledge graph, and form a fault entity generalized name dataset from the generalized names of each fault entity; construct a fine-tuned language model, and train the fine-tuned language model together using the fault entity original name dataset and the fault entity generalized name dataset to obtain an entity semantic indexing model that can calculate word vectors for both fault entity original names and fault entity generalized names;
[0011] S4: Calculate the word vectors corresponding to the original names of each fault entity in the fault entity original name dataset through the entity semantic indexing model, and store the relationship between the original names of fault entities and their word vectors in the vector database.
[0012] S5: Obtain the new entity. First, extract the original name of the new entity. Input the extracted original name of the new entity into the entity semantic index model to calculate its word vector, and obtain the word vector of the new entity. Input the word vector of the new entity into the vector database for querying and obtain the original name of the fault entity with the smallest distance value. If the smallest distance value is less than the preset threshold, it is considered that the new entity and the fault entity corresponding to the original name of the fault entity are the same, and the relevant information is updated in the generalized automotive fault knowledge graph; otherwise, the entity and related information are added to the generalized automotive fault knowledge graph.
[0013] Furthermore, in step S1, the fault entities and the relationships between multiple fault entities are extracted from the initial vehicle fault knowledge through the entity recognition model.
[0014] Furthermore, in step S3, the language model is fine-tuned using the SimCSE model.
[0015] Furthermore, in step S4, the vector database into which the relationship between the original name of the faulty entity and its word vector is stored is the Milvus vector database.
[0016] Further, in step S5, if the newly added entity is the same as the fault entity corresponding to the original name of the fault entity, the correspondence between the original name of the newly added entity and the fault entity with the same name is added to the generalized automotive fault knowledge graph; if the newly added entity is not the same as the fault entity corresponding to the original name of the fault entity, the name of the newly added entity is generalized using a word vector model to obtain the generalized name of the newly added entity, and the generalized name of the newly added entity, the original name of the newly added entity, and the relationship between the generalized name of the newly added entity and the original name of the newly added entity are stored in the generalized automotive fault knowledge graph, and the word vector corresponding to the original name of the newly added entity is calculated through the entity semantic index model, and the relationship between the original name of the newly added entity and its word vector is stored in the vector database.
[0017] Furthermore, in step S5, the new knowledge is the user input information, and the W2NER model is used to extract entities from the user input information; when the user performs entity retrieval, the original names of one or more corresponding fault entities with the smallest distance value are returned to the user, and the fault entities corresponding to the original names of the fault entities are used to obtain knowledge from the generalized automotive fault knowledge graph.
[0018] Furthermore, it also includes:
[0019] S6: Extract new automotive fault knowledge from user feedback, perform correlation analysis with existing automotive fault knowledge, establish new semantic links, and update the correlation in the generalized automotive fault knowledge graph, while also updating the vector database.
[0020] Furthermore, it also includes:
[0021] S7: When the number of newly added entities in the generalized automotive fault knowledge graph or the time since the last update of the entity semantic index model exceeds a preset threshold, step S3 is executed to update the entity semantic index model.
[0022] This invention also provides a system for constructing an automotive fault knowledge graph, comprising:
[0023] The initial vehicle fault knowledge graph generation module is used to acquire initial vehicle fault knowledge, extract fault entities and the relationships between multiple fault entities from the initial vehicle fault knowledge, and generate an initial vehicle fault knowledge graph.
[0024] The generalized automotive fault knowledge graph generation module is used to generalize the original names of fault entities in the initial automotive fault knowledge graph using a word vector model to obtain generalized names of fault entities. The generalized names of fault entities and the original names of fault entities are stored together in the initial automotive fault knowledge graph, and the relationship between the generalized names of fault entities and the original names of fault entities is established to generate a generalized automotive fault knowledge graph.
[0025] The entity semantic indexing model generation module is used to obtain the original names of each fault entity from the generalized automotive fault knowledge graph, and to form a fault entity original name dataset; to obtain the generalized names of each fault entity from the generalized automotive fault knowledge graph, and to form a fault entity generalized name dataset; to construct a fine-tuned language model, and to train the fine-tuned language model together with the fault entity original name dataset and the fault entity generalized name dataset to obtain an entity semantic indexing model that can calculate the word vectors of both the fault entity original names and the fault entity generalized names.
[0026] The vector database generation module is used to calculate the word vectors corresponding to the original names of each fault entity in the fault entity original name dataset through the entity semantic indexing model, and to store the relationship between the original names of fault entities and their word vectors into the vector database.
[0027] The update module is used to obtain new entities. First, it extracts the original name of the new entity and inputs it into the entity semantic index model to calculate its word vector, thus obtaining the word vector of the new entity. Then, it inputs the word vector of the new entity into the vector database for querying to obtain the original name of the fault entity with the smallest distance value. If the smallest distance value is less than a preset threshold, it is considered that the new entity and the fault entity corresponding to the original name of the fault entity are the same, and the relevant information is updated in the generalized automotive fault knowledge graph; otherwise, the entity and related information are added to the generalized automotive fault knowledge graph.
[0028] Furthermore, the automotive fault knowledge graph construction system also includes:
[0029] The knowledge feedback module extracts new automotive fault knowledge from user feedback, performs correlation analysis with existing automotive fault knowledge, establishes new semantic links, and updates the correlation in the generalized automotive fault knowledge graph, while also updating the vector database.
[0030] The revamp module is used to determine whether the number of newly added entities in the generalized automotive fault knowledge graph or the time since the last update of the entity semantic index model exceeds a preset threshold. If the number of newly added entities in the generalized automotive fault knowledge graph or the time since the last update of the entity semantic index model exceeds the preset threshold, the entity semantic index model is updated.
[0031] The beneficial effects of this invention are:
[0032] This invention improves the generalization ability of knowledge graphs by generalizing fault entities, solving the problem of matching synonyms and near-synonyms in entity name descriptions, achieving "learning by analogy," reducing the matching difficulty, and training an entity semantic indexing model using a dataset of fault entity original names (composed of the original names of each fault entity obtained from the generalized automotive fault knowledge graph) and a dataset of fault entity generalized names (composed of the generalized names of each fault entity obtained from the generalized automotive fault knowledge graph). The entity semantic indexing model can predict the word vectors of fault entities, and then retrieve one or more fault entities with the smallest distance value after searching the vector database using the word vectors calculated by the entity semantic indexing model. This eliminates the need for similarity calculations, simplifying the process and achieving high efficiency and accuracy. The entity semantic indexing model is trained from the fault entity original name dataset and the fault entity generalized name dataset, exhibiting strong generalization ability and improved accuracy. Attached Figure Description
[0033] Figure 1 This is a flowchart of the automotive fault knowledge graph construction method of the present invention.
[0034] Figure 2 This is a schematic diagram of the automotive fault knowledge graph construction system of the present invention. Detailed Implementation
[0035] The embodiments of the present invention will be described below with reference to the accompanying drawings and preferred embodiments. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be understood that the preferred embodiments are only for illustrating the present invention and not for limiting the scope of protection of the present invention.
[0036] The relevant terms in this invention embodiment are explained as follows:
[0037] Automotive fault knowledge: includes basic vehicle information, vehicle owner's manual, repair records, maintenance records, vehicle fault diagnosis solutions, vehicle parts list, vehicle condition records, fault records, etc.;
[0038] Automotive fault knowledge graph: includes fault ontology and knowledge links;
[0039] Fault ontology: includes a description of the fault ontology concept, ontology attributes, and ontology relationships; the fault ontology concept of this invention includes brand, vehicle series, vehicle model, component, fault, root cause, phenomenon, solution, repair, maintenance, vehicle, and vehicle condition;
[0040] Specifically:
[0041] Brand: Indicates the car brand;
[0042] Car series: Indicates a series of car models;
[0043] Vehicle model: refers to the model of a car, specifically the different styles or versions of the car launched based on different configurations and configuration combinations;
[0044] Components: These refer to automotive body modules or parts, such as VCU, ECU, engine, tires, and central control screen.
[0045] Faults: These indicate various problems and malfunctions that may occur in a car, such as difficulty starting, cooling system failure, brake failure, abnormal noise from the brake discs, and a frozen central control screen.
[0046] Root cause: refers to the fundamental reason for a car malfunction, specifically the problem related to the failure, damage, or alteration of car parts or systems. Root causes can be categorized into various types, such as mechanical failures, electrical failures, and electronic control system failures.
[0047] Phenomenon: This refers to a car malfunction, specifically a sudden failure, abnormal operation, or inability to drive normally during the use of the car. Common malfunctions include difficulty starting the engine, shaking while driving, excessive fuel consumption, and brake failure.
[0048] Solution: Describes the methods and techniques for eliminating the root causes of car malfunctions;
[0049] Repair: Indicates vehicle malfunction and repair information;
[0050] Maintenance: This section indicates information about the car's routine maintenance.
[0051] Vehicles: Indicates either gasoline-powered vehicles or new energy vehicles;
[0052] Vehicle condition: This indicates the vehicle's condition information, including but not limited to vehicle status, charging status, refueling status, operating mode, vehicle speed, cumulative mileage, SOC, total voltage, total current, gear position, engine status, and drive motor status.
[0053] Ontology attributes are further descriptions of ontology features; for example, specific attributes of "brand" include name and code; specific attributes of "vehicle" include VIN, model, length, width, height, and wheelbase.
[0054] Ontology relationships include: vehicle model, vehicle component, vehicle fault, submodule, root cause of fault, fault phenomenon, repair case, repair suggestion, repair record, maintenance record, vehicle condition record, fault record, and fault signal; specifically: the vehicle model represents the relationship between brand and vehicle model, with the brand entity pointing to the vehicle model entity, establishing the relationship between brand and vehicle model, allowing for easy identification of vehicle model based on brand or brand based on vehicle model; the vehicle component represents the relationship between vehicle model and component, with the vehicle model entity pointing to the component entity, establishing the relationship between vehicle model and component, describing which components are under each vehicle model; the vehicle fault represents the relationship between component and fault, with the component entity pointing to the fault entity, establishing the relationship between component and fault. Each component describes the possible faults it may experience; each submodule represents the relationship between components, with parent component entities pointing to child component entities, establishing the relationship between components and describing the composition of each component's subcomponents; each root cause represents the relationship between faults and root causes, with fault entities pointing to root cause entities, establishing the relationship between faults and root causes, and describing the factors that may cause each fault; each fault phenomenon represents the relationship between faults and phenomena, with fault entities pointing to phenomenon entities, establishing the relationship between faults and phenomena, describing the specific phenomena of each fault, which can be used to find specific fault entities through fault phenomenon descriptions; each repair case represents the relationship between faults and repairs, with fault entities pointing to repair entities, establishing the relationship between faults and repairs. Each fault description provides a repair case, which can be used to recommend relevant fault repair cases based on user-searched fault content. The fault repair suggestions represent the relationship between faults and solutions, pointing from the fault entity to the solution entity, establishing the relationship between faults and solutions, and describing the specific solutions or repair suggestions for each fault. The component repair suggestions represent the relationship between components and solutions, pointing from the component entity to the solution entity, establishing the relationship between components and solutions, and describing the component's repair methods or repair suggestions. The repair records represent the relationship between vehicles and repairs, pointing from the vehicle entity to the repair entity, constructing the relationship between vehicles and repairs, and describing the vehicle's specific repair information. The maintenance records represent the relationship between vehicles and maintenance, pointing from the vehicle entity to the maintenance entity. The system comprises three main components: a vehicle entity, a vehicle condition entity, and a fault record. The vehicle condition entity represents the relationship between the vehicle and its maintenance, describing the specific maintenance information of the vehicle. The fault record represents the relationship between the vehicle and its fault, describing the specific fault information of the vehicle. The fault signal represents the relationship between the vehicle condition and its fault, describing the fault information corresponding to the fault signal. The knowledge link is constructed using the fault ontology, ontology attributes, and ontology relationships, specifically including vehicle fault analysis, fault detection, fault warning, and fault consultation.
[0055] Knowledge Link: Constructed using the aforementioned fault ontology, ontology attributes, and ontology relationships, specifically including vehicle fault analysis, fault detection, fault early warning, and fault consultation;
[0056] Entity linking: is the process of standardizing and linking entities from different data sources;
[0057] Knowledge merging: This involves integrating knowledge points from different data sources.
[0058] Knowledge processing involves extracting faulty entities and entity relationships from different data sources and then performing knowledge reasoning and supplementation.
[0059] The automotive fault knowledge organization types include structured, semi-structured, and unstructured data. The structured data is transformed into entities, entity attribute values, and entity relationship data from one or more data sources through the establishment of fault ontology mapping, ontology attribute mapping, and ontology relationship mapping, and synchronized to the automotive fault knowledge graph in real time. The semi-structured and unstructured data are labeled with sample data, and the sample data is used to train automotive fault entity recognition models, automotive fault entity attribute extraction models, and automotive fault entity relationship extraction models, respectively.
[0060] like Figure 1 As shown, a preferred embodiment of the present invention provides a method for constructing an automotive fault knowledge graph, comprising:
[0061] S1: Obtain initial vehicle fault knowledge, extract fault entities and the relationships between multiple fault entities from the initial vehicle fault knowledge, and generate an initial vehicle fault knowledge graph.
[0062] S2: Generalize the original names of fault entities in the initial vehicle fault knowledge graph using a word vector model to obtain generalized names of fault entities. Store the generalized names and original names of fault entities in the initial vehicle fault knowledge graph and establish the relationship between the generalized names and original names of fault entities to generate a generalized vehicle fault knowledge graph.
[0063] S3: Obtain the original names of each fault entity from the generalized automotive fault knowledge graph, and form a fault entity original name dataset from the original names of each fault entity; obtain the generalized names of each fault entity from the generalized automotive fault knowledge graph, and form a fault entity generalized name dataset from the generalized names of each fault entity; construct a fine-tuned language model, and train the fine-tuned language model together using the fault entity original name dataset and the fault entity generalized name dataset to obtain an entity semantic indexing model that can calculate word vectors for both fault entity original names and fault entity generalized names;
[0064] S4: Calculate the word vectors corresponding to the original names of each fault entity in the fault entity original name dataset through the entity semantic indexing model, and store the relationship between the original names of fault entities and their word vectors in the vector database.
[0065] S5: Obtain the new entity. First, extract the original name of the new entity. Input the extracted original name of the new entity into the entity semantic index model to calculate its word vector, thus obtaining the word vector of the new entity. Input the word vector of the new entity into the vector database for querying. Obtain the original name of the fault entity with the smallest distance value. If the smallest distance value is less than a preset threshold, it is considered that the new entity and the fault entity corresponding to the original name of the fault entity are the same, and the relevant information is updated in the generalized automotive fault knowledge graph. Otherwise, add the entity and related information to the generalized automotive fault knowledge graph. This step realizes the acquisition of new knowledge and the updating of the generalized automotive fault knowledge graph and vector database based on the new knowledge.
[0066] S6: Extract new automotive fault knowledge from user feedback, perform correlation analysis with existing automotive fault knowledge, establish new semantic links, and update the correlation in the generalized automotive fault knowledge graph, while also updating the vector database; realize the acquisition of user feedback information and knowledge correction based on user feedback information.
[0067] S7: When the number of newly added entities in the generalized automotive fault knowledge graph or the time since the last update of the entity semantic index model exceeds a preset threshold, step S3 is executed to update the entity semantic index model. The preset threshold is set by the user.
[0068] This embodiment improves the generalization ability of the knowledge graph by generalizing fault entities, solving the problem of matching synonyms and near-synonyms in entity name descriptions, achieving "learning by analogy," reducing the matching difficulty, and training an entity semantic indexing model using a dataset of original fault entity names obtained from the generalized automotive fault knowledge graph and a dataset of generalized fault entity names obtained from the generalized automotive fault knowledge graph. The entity semantic indexing model can predict the word vectors of fault entities, and then use the word vectors calculated by the entity semantic indexing model to retrieve one or more fault entities with the smallest distance value after searching the vector database. This eliminates the need for similarity calculations, simplifying the process and improving efficiency and accuracy. The entity semantic indexing model is trained from the dataset of original fault entity names and the dataset of generalized fault entity names, demonstrating strong generalization ability and improved accuracy.
[0069] In addition, the system updates the generalized automotive fault knowledge graph and vector database based on new knowledge, updates the knowledge graph based on user feedback, and updates the entity semantic index model when the number of new entities in the generalized automotive fault knowledge graph or the time since the last update of the entity semantic index model exceeds a preset threshold. This constructs a feedback and update mechanism for the automotive fault knowledge graph, enabling automatic updates of the automotive fault ontology, fault entities, and knowledge links, as well as automatic optimization and correction of the automotive fault knowledge graph. This provides more accurate knowledge for scenarios such as automotive fault analysis, fault detection, fault warning, and fault consultation.
[0070] Specifically, in step S1, the entity recognition model extracts fault entities and the relationships between multiple fault entities from the initial vehicle fault knowledge.
[0071] In step S3, the language model is fine-tuned using the SimCSE model.
[0072] In step S4, the relationship between the original name of the faulty entity and its word vector is stored in the Milvus vector database.
[0073] In step S5, if the newly added entity is the same as the fault entity corresponding to the original name of the fault entity, the correspondence between the original name of the newly added entity and the fault entity with the same name is added to the generalized automotive fault knowledge graph; if the newly added entity is not the same as the fault entity corresponding to the original name of the fault entity, the name of the newly added entity is generalized using a word vector model to obtain the generalized name of the newly added entity. The generalized name of the newly added entity, the original name of the newly added entity, and the relationship between the generalized name of the newly added entity and the original name of the newly added entity are stored in the generalized automotive fault knowledge graph. The word vector corresponding to the original name of the newly added entity is calculated through the entity semantic index model, and the relationship between the original name of the newly added entity and its word vector is stored in the vector database.
[0074] In step S5, the new knowledge is the user input information. The W2NER model is used to extract entities from the user input information. When the user performs entity retrieval, the original names of one or more corresponding fault entities with the smallest distance value are returned to the user. The fault entities corresponding to the original names of the fault entities are used to obtain knowledge from the generalized automotive fault knowledge graph.
[0075] like Figure 2 As shown, this embodiment of the invention also provides a vehicle fault knowledge graph construction system, including:
[0076] The initial vehicle fault knowledge graph generation module is used to acquire initial vehicle fault knowledge, extract fault entities and the relationships between multiple fault entities from the initial vehicle fault knowledge, and generate an initial vehicle fault knowledge graph.
[0077] The generalized automotive fault knowledge graph generation module is used to generalize the original names of fault entities in the initial automotive fault knowledge graph using a word vector model to obtain generalized names of fault entities. The generalized names of fault entities and the original names of fault entities are stored together in the initial automotive fault knowledge graph, and the relationship between the generalized names of fault entities and the original names of fault entities is established to generate a generalized automotive fault knowledge graph.
[0078] The entity semantic indexing model generation module is used to obtain the original names of each fault entity from the generalized automotive fault knowledge graph, and to form a fault entity original name dataset; to obtain the generalized names of each fault entity from the generalized automotive fault knowledge graph, and to form a fault entity generalized name dataset; to construct a fine-tuned language model, and to train the fine-tuned language model together with the fault entity original name dataset and the fault entity generalized name dataset to obtain an entity semantic indexing model that can calculate the word vectors of both the fault entity original names and the fault entity generalized names.
[0079] The vector database generation module is used to calculate the word vectors corresponding to the original names of each fault entity in the fault entity original name dataset through the entity semantic indexing model, and to store the relationship between the original names of fault entities and their word vectors into the vector database.
[0080] The update module is used to obtain new entities. First, it extracts the original name of the new entity and inputs it into the entity semantic index model to calculate its word vector, thus obtaining the word vector of the new entity. Then, it inputs the word vector of the new entity into the vector database for querying to obtain the original name of the fault entity with the smallest distance value. If the smallest distance value is less than a preset threshold, it is considered that the new entity and the fault entity corresponding to the original name of the fault entity are the same, and the relevant information is updated in the generalized automotive fault knowledge graph; otherwise, the entity and related information are added to the generalized automotive fault knowledge graph.
[0081] The knowledge feedback module is used to obtain user feedback information and correct knowledge based on the user feedback information. Specifically, it extracts new car fault knowledge from user feedback information, performs correlation analysis with existing car fault knowledge, establishes new semantic links, and updates the correlation in the generalized car fault knowledge graph, while also updating the vector database.
[0082] The revamp module is used to determine whether the number of newly added entities in the generalized automotive fault knowledge graph or the time since the last update of the entity semantic index model exceeds a preset threshold. If the number of newly added entities in the generalized automotive fault knowledge graph or the time since the last update of the entity semantic index model exceeds the preset threshold, the entity semantic index model is updated.
[0083] Specifically, the initial automotive fault knowledge graph generation module includes a knowledge extraction unit, a knowledge storage unit, and a knowledge management unit. The knowledge extraction unit is used to obtain structured, semi-structured, and unstructured knowledge from one or more data sources, and then extract and fuse automotive fault knowledge before writing it into the knowledge storage unit. The knowledge storage unit mainly uses a graph database to store triplet fault knowledge. The knowledge management unit mainly manages the metadata information of automotive fault knowledge.
[0084] In summary, this invention provides a method for constructing an automotive fault knowledge graph. By generalizing fault entities, it improves the generalization ability of the knowledge graph, solves the matching problem of synonyms and near-synonyms in entity name descriptions, achieves "learning by analogy," reduces the matching difficulty, and trains an entity semantic indexing model using a dataset of original fault entity names obtained from the generalized automotive fault knowledge graph and a dataset of generalized fault entity names obtained from the generalized automotive fault knowledge graph. The entity semantic indexing model can predict the word vectors of fault entities, and then retrieves one or more fault entities with the smallest distance value after searching the vector database using the word vectors calculated by the entity semantic indexing model. This eliminates the need for similarity calculations, simplifying the process and improving efficiency and accuracy. The entity semantic indexing model is trained from the dataset of original fault entity names and the dataset of generalized fault entity names, exhibiting strong generalization ability and improved accuracy. Furthermore, a feedback and update mechanism for the automotive fault knowledge graph is constructed by updating the generalized automotive fault knowledge graph and vector database based on new knowledge, updating the knowledge graph based on user feedback, and updating the entity semantic index model when the number of newly added entities in the generalized automotive fault knowledge graph or the time since the last update of the entity semantic index model exceeds a preset threshold. This mechanism enables automatic updating of the automotive fault ontology, fault entities, and knowledge links, and automatic optimization and correction of the automotive fault knowledge graph, providing more accurate knowledge for scenarios such as automotive fault analysis, fault detection, fault early warning, and fault consultation. This invention also provides an automotive fault knowledge graph construction system based on the above method.
[0085] The above embodiments are merely preferred embodiments provided to fully illustrate the present invention, and the scope of protection of the present invention is not limited thereto. Equivalent substitutions or modifications made by those skilled in the art based on the present invention are all within the scope of protection of the present invention.
Claims
1.A method for constructing an automobile fault knowledge graph, characterized in that, include: S1: Obtain initial vehicle fault knowledge, extract fault entities and the relationships between multiple fault entities from the initial vehicle fault knowledge, and generate an initial vehicle fault knowledge graph. S2: Generalize the original names of fault entities in the initial vehicle fault knowledge graph using a word vector model to obtain generalized names of fault entities. Store the generalized names and original names of fault entities in the initial vehicle fault knowledge graph and establish the relationship between the generalized names and original names of fault entities to generate a generalized vehicle fault knowledge graph. S3: Obtain the original names of each fault entity from the generalized automotive fault knowledge graph, and form a fault entity original name dataset from the original names of each fault entity; obtain the generalized names of each fault entity from the generalized automotive fault knowledge graph, and form a fault entity generalized name dataset from the generalized names of each fault entity. A fine-tuned language model is constructed, and the fine-tuned language model is trained together with the original name dataset and the generalized name dataset of faulty entities to obtain an entity semantic indexing model that can calculate word vectors for both the original name and the generalized name of faulty entities. S4: Calculate the word vectors corresponding to the original names of each fault entity in the fault entity original name dataset through the entity semantic indexing model, and store the relationship between the original names of fault entities and their word vectors in the vector database. S5: Obtain the new entity. First, extract the original name of the new entity. Input the extracted original name of the new entity into the entity semantic index model to calculate its word vector and obtain the word vector of the new entity. Input the word vector of the new entity into the vector database for querying and obtain the original name of the fault entity with the smallest distance value. If the smallest distance value is less than the preset threshold, it is considered that the new entity and the fault entity corresponding to the original name of the fault entity are the same, and the relevant information is updated to the generalized automotive fault knowledge graph. Otherwise, add entities and related information to the generalized automotive fault knowledge graph. 2.The automobile fault knowledge graph construction method according to claim 1, characterized in that, In step S1, the entity recognition model extracts fault entities and the relationships between multiple fault entities from the initial vehicle fault knowledge. 3.The automobile fault knowledge graph construction method according to claim 1, characterized in that, In step S3, the language model is fine-tuned using the SimCSE model. 4.The automobile fault knowledge graph construction method according to claim 1, characterized in that, In step S4, the relationship between the original name of the faulty entity and its word vector is stored in the Milvus vector database. 5.The automobile fault knowledge graph construction method according to claim 1, characterized in that, In step S5, if the newly added entity is the same as the fault entity corresponding to the original name of the fault entity, the correspondence between the original name of the newly added entity and the fault entity with the same name is added to the generalized automotive fault knowledge graph; if the newly added entity is not the same as the fault entity corresponding to the original name of the fault entity, the name of the newly added entity is generalized using a word vector model to obtain the generalized name of the newly added entity. The generalized name of the newly added entity, the original name of the newly added entity, and the relationship between the generalized name of the newly added entity and the original name of the newly added entity are stored in the generalized automotive fault knowledge graph. The word vector corresponding to the original name of the newly added entity is calculated through the entity semantic index model, and the relationship between the original name of the newly added entity and its word vector is stored in the vector database. 6.The automobile fault knowledge graph construction method according to claim 1, characterized in that, In step S5, the new knowledge is the user input information. The W2NER model is used to extract entities from the user input information. When the user performs entity retrieval, the original names of one or more corresponding fault entities with the smallest distance value are returned to the user. The fault entities corresponding to the original names of the fault entities are used to obtain knowledge from the generalized automotive fault knowledge graph. 7.The automobile fault knowledge graph construction method according to claim 1, characterized in that, Also includes: S6: Extract new automotive fault knowledge from user feedback, perform correlation analysis with existing automotive fault knowledge, establish new semantic links, and update the correlation in the generalized automotive fault knowledge graph, while also updating the vector database. 8.The method of claim 1, wherein, Also includes: S7: When the number of newly added entities in the generalized automotive fault knowledge graph or the time since the last update of the entity semantic index model exceeds a preset threshold, step S3 is executed to update the entity semantic index model. 9.A system for constructing an automobile failure knowledge graph, characterized by comprising: include: The initial vehicle fault knowledge graph generation module is used to acquire initial vehicle fault knowledge, extract fault entities and the relationships between multiple fault entities from the initial vehicle fault knowledge, and generate an initial vehicle fault knowledge graph. The generalized automotive fault knowledge graph generation module is used to generalize the original names of fault entities in the initial automotive fault knowledge graph using a word vector model to obtain generalized names of fault entities. The generalized names of fault entities and the original names of fault entities are stored together in the initial automotive fault knowledge graph, and the relationship between the generalized names of fault entities and the original names of fault entities is established to generate a generalized automotive fault knowledge graph. The entity semantic index model generation module is used to obtain the original names of each fault entity from the generalized automotive fault knowledge graph, and to form a fault entity original name dataset from the original names of each fault entity; and to obtain the generalized names of each fault entity from the generalized automotive fault knowledge graph, and to form a fault entity generalized name dataset from the generalized names of each fault entity. A fine-tuned language model is constructed, and the fine-tuned language model is trained together with the original name dataset and the generalized name dataset of faulty entities to obtain an entity semantic indexing model that can calculate word vectors for both the original name and the generalized name of faulty entities. The vector database generation module is used to calculate the word vectors corresponding to the original names of each fault entity in the fault entity original name dataset through the entity semantic indexing model, and to store the relationship between the original names of fault entities and their word vectors into the vector database. The update module is used to obtain new entities. First, the original name of the new entity is extracted. The extracted original name of the new entity is input into the entity semantic index model to calculate its word vector, and the word vector of the new entity is obtained. The word vector of the new entity is input into the vector database for querying to obtain the original name of the fault entity with the smallest distance value. If the smallest distance value is less than a preset threshold, it is considered that the new entity and the fault entity corresponding to the original name of the fault entity are the same, and the relevant information is updated to the generalized automotive fault knowledge graph. Otherwise, add entities and related information to the generalized automotive fault knowledge graph. 10.The automobile fault knowledge graph construction system according to claim 9, characterized in that, Also includes: The knowledge feedback module is used to extract new automotive fault knowledge from user feedback information, perform correlation analysis with existing automotive fault knowledge, establish new semantic links, and update the correlation in the generalized automotive fault knowledge graph, while also updating the vector database. The revamp module is used to determine whether the number of newly added entities in the generalized automotive fault knowledge graph or the time since the last update of the entity semantic index model exceeds a preset threshold. If the number of newly added entities in the generalized automotive fault knowledge graph or the time since the last update of the entity semantic index model exceeds the preset threshold, the entity semantic index model is updated.