CN115048532A - Intelligent question-answering robot for automobile maintenance scene based on knowledge graph and design method - Google Patents

Abstract

The invention discloses an intelligent question answering robot and a design method for automobile maintenance scenarios based on a knowledge graph, comprising the following steps: constructing a knowledge graph for automobile maintenance; Extract the node attributes and node relationship information in the knowledge graph from the semi-structured and semi-structured maintenance data; store the extracted data in the graph database to form the final usable auto maintenance knowledge graph; take the constructed auto maintenance knowledge graph as the knowledge base , develop intelligent robots. The invention can help the front-line maintenance technicians to quickly obtain the maintenance method and related information of the fault.

Description

Intelligent question answering robot and design method for automobile maintenance scene based on knowledge graph

technical field

The invention belongs to the technical field of Internet of Vehicles knowledge question answering, and in particular relates to an intelligent question answering robot for automobile maintenance scenarios based on a knowledge graph and a design method thereof.

Background technique

At present, in the automobile maintenance scene, the relevant maintenance technicians are required to be able to quickly and accurately locate the faulty parts of the maintenance vehicles and propose relevant solutions. However, because the car has many system components and sensors, and there are many car series under one brand, and each car series has many different styles, it is difficult for general maintenance technicians to quickly locate the cause of the fault, and manual inquiry is required. Maintenance information, often calling for remote support from experts in the field, which leads to a long maintenance process, low maintenance efficiency, and affects the user experience; at the same time, a large number of faults have occurred, but these fault record information Different forms of storage, such as image format, XML format, etc., are separated from each other and fail to form an effective knowledge fusion; in addition, the time of domain experts is occupied by a large number of maintenance technicians, and the efficiency of maintenance and telephone communication is low.

Artificial intelligence has matured applications in many fields, bringing great value. As the third generation of artificial intelligence technology, knowledge graph has been widely used in many fields such as intelligent question answering, search recommendation, and intelligent reasoning. In this context, we use knowledge graph technology to establish relationships with large amounts of data, mine knowledge, integrate, manage and utilize knowledge resources, establish a knowledge sharing mechanism, and enhance the value of data.

At present, there are existing solutions to provide a question and answer system for maintenance scenarios, which uses traditional text search, database data query and other technologies to achieve data query and acquisition functions. This method faces many shortcomings. First, the task-based data acquisition method is used, which is inefficient and takes a long time for users to wait. Second, the accuracy of data query is not high. Finally, data maintenance is difficult.

SUMMARY OF THE INVENTION

In order to solve the above problems existing in the prior art, the present invention provides a knowledge graph-based intelligent question answering robot for automobile maintenance scenarios and a design method thereof, which can help front-line maintenance technicians to quickly obtain fault maintenance methods and related information. With the help of the maintenance knowledge graph, the present invention can effectively precipitate and connect maintenance data to form knowledge; establish an intelligent question answering system based on the knowledge graph, and use NLP technology to analyze the original unstructured and semi-structured data to construct The knowledge relationship of the knowledge graph; the introduction of an intelligent question and answer robot, using a multi-round question and answer engine, pushes accurate query information and search results to users, improving maintenance efficiency and user experience.

The purpose of this invention is to realize through the following technical solutions:

As an aspect of the present invention, a kind of intelligent question answering robot based on knowledge graph for automobile maintenance scene is provided, including:

The maintenance knowledge graph module, which combines the scene characteristics of the maintenance field, establishes the knowledge graph ontology, and extracts the node attributes and node-related information in the knowledge graph ontology and stores it in the graph database;

The intelligent question answering module is in communication connection with the maintenance knowledge graph module, performs language analysis on the input voice or text, and performs query retrieval from the maintenance knowledge graph module, and sorts and outputs the query retrieval results.

Further, the maintenance knowledge graph module includes:

The knowledge graph ontology unit, which includes the nodes of vehicle maintenance and the node relationship between the nodes; the nodes include: fault phenomenon, fault code, parts, maintenance cases, and vehicle series;

A knowledge extraction unit, which extracts relevant information such as node attributes and node relationships from the knowledge graph ontology unit;

The graph database unit is used for the knowledge information extracted by the knowledge extraction unit to form the final maintenance knowledge graph.

Furthermore, the graph database unit converts the data obtained by the knowledge extraction unit into an input format supported by the graph database, and stores the data in the graph database.

Still further, the graph database is a database for semantic query using a graph structure, which uses nodes, edges and attributes to represent and store data.

Further, the intelligent question answering module includes:

A natural language parsing unit, which converts single or multiple rounds of text data input by the user into one or more graph query statements and corresponding parameters;

a recall unit, which inputs the graph query statement generated by the natural language parsing unit into the maintenance knowledge graph module for retrieval and query, receives the returned result of the query, and performs preliminary screening of the results;

a sorting unit, which performs similarity calculation and scoring on the results after the preliminary screening of the recall unit, sorts the results according to the scores, and selects the results with the top ranking and sends them to the result presentation unit;

The result presentation unit, which presents different forms according to the types of results sent by the sorting unit.

Further, the sorting unit includes a questioning sub-unit, which is used to calculate whether there is a difference in the attribute value of each node or the attribute value of a hop associated entity in the recalled knowledge graph entity node set; if there is a difference, there will be a difference in the attribute value. As the content of the question, it is pushed to the user, and according to the user's answer to the question, the corresponding answer is screened and presented to the next step.

Further, the presentation type of the result presentation unit includes text, picture, video, and audio form.

As another aspect of the present invention, a method for designing an intelligent question answering robot in a vehicle maintenance scene based on a knowledge graph is provided, comprising the following steps:

S1. Build a car maintenance knowledge map:

S11. Ontology design: Design the auto maintenance knowledge map ontology based on the scene characteristics of the auto maintenance field;

S12. Knowledge extraction: extract node attributes and node relationship information in the knowledge graph from structured, unstructured and semi-structured maintenance data;

S13. Knowledge storage: store the extracted data in the graph database to form the final usable auto maintenance knowledge graph;

S2. Intelligent question-answering robot development: develop intelligent robots with the constructed auto maintenance knowledge graph as the knowledge base.

Further, in the ontology design of step S11 , which concepts exist in the knowledge graph and the existing relationships between these concepts are described through ontology.

Further, in the knowledge extraction in the step S12, if the data for saving knowledge is structured data, a template is written, and triples are extracted from it; if the data for saving knowledge is unstructured data, then a customized programming language is used to write and parse. Scripts, parsing raw data, structuring it.

Further, in the knowledge extraction in step S12, if the data storing the knowledge is unstructured data, entities and relationships are extracted from the unstructured data by training entity extraction and relationship extraction algorithm models.

Further, in the knowledge storage in step S13, the data obtained in the knowledge extraction step is converted into an input format supported by the graph database, and stored in the graph database.

Further, the step S2 intelligent question answering robot development includes:

S21. Natural language parsing: The robot conversation engine first understands the user's input intent, and converts the single or multiple rounds of text data input by the user into one or more graph query statements and corresponding parameters;

S22. Recall: input the graph query statement generated in the previous step into the knowledge graph, and receive the returned results of the query, and perform preliminary screening of the results;

S23. Sorting: For the recalled results, use the semantic engine to calculate and score the similarity, sort according to the scores, and select the results with the highest ranking;

S24. Answer presentation: different forms of presentation are performed according to the type of the returned result.

Further, in the natural language parsing of step S21, the robot conversation engine understands the user's intention through an entity linking algorithm.

Further, the step S23 sorting is provided with a questioning function: in the set of knowledge graph entity nodes of the recalled knowledge graph, whether there is a difference in the attribute value of each node or the attribute value of a hop associated entity; if there is a difference, there will be a difference. Attributes are used as the content of the question, and the user is asked, and according to the user's answer to the question, the corresponding answer is screened and presented to the next step.

Compared with the prior art, the present invention has the following advantages:

(1) Compared with the traditional plain text data method, the question answering system based on knowledge graph has more advantages in data correlation. The degree of semantic understanding is a core indicator of the question answering system. In the knowledge graph, all knowledge points are connected by edges with semantic information. From a question sentence to the knowledge points associated with the knowledge graph, a large number of knowledge points are undoubtedly used. The edge of semantic information, this kind of related information plays an important role in intelligent question answering system.

(2) At the same time, the question answering system based on knowledge graph has more advantages in data accuracy. The knowledge in the knowledge graph is marked or formatted and parsed by professionals, which ensures high data accuracy.

(3) Finally, the question answering system based on knowledge graph has more advantages in data structured search. Compared with the traditional knowledge search method, the knowledge graph-based search can search for "things", and at the same time display various information such as attributes and relationships of the thing.

Description of drawings

Fig. 1 is a kind of knowledge graph-based intelligent question answering robot design process in automobile maintenance scene according to the embodiment of the present invention;

2 is a schematic diagram of a maintenance knowledge graph ontology according to an embodiment of the present invention;

FIG. 3 is a flow chart of using an intelligent question answering robot in a vehicle maintenance scene according to an embodiment of the present invention.

Detailed ways

The present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention, but not to limit the present invention.

An intelligent question answering robot based on knowledge graph for automobile maintenance scene, including:

The maintenance knowledge graph module, which combines the scene characteristics of the maintenance field, establishes the knowledge graph ontology, and extracts the node attributes and node-related information in the knowledge graph ontology and stores it in the graph database;

The intelligent question answering module is in communication connection with the maintenance knowledge graph module, performs language analysis on the input voice or text, and performs query retrieval from the maintenance knowledge graph module, and sorts and outputs the query retrieval results.

Further, the maintenance knowledge graph module includes:

The knowledge graph ontology unit, which includes the nodes of vehicle maintenance and the node relationship between the nodes; the nodes include: fault phenomenon, fault code, parts, maintenance cases, and vehicle series;

A knowledge extraction unit, which extracts relevant information such as node attributes and node relationships from the knowledge graph ontology unit;

The graph database unit is used for the knowledge information extracted by the knowledge extraction unit to form the final maintenance knowledge graph.

Furthermore, the graph database unit converts the data obtained by the knowledge extraction unit into an input format supported by the graph database, and stores the data in the graph database.

Still further, the graph database is a database for semantic query using a graph structure, which uses nodes, edges and attributes to represent and store data.

Further, the intelligent question answering module includes:

A natural language parsing unit, which converts single or multiple rounds of text data input by the user into one or more graph query statements and corresponding parameters;

a recall unit, which inputs the graph query statement generated by the natural language parsing unit into the maintenance knowledge graph module for retrieval and query, receives the returned result of the query, and performs preliminary screening of the results;

a sorting unit, which performs similarity calculation and scoring on the results after the preliminary screening of the recall unit, sorts the results according to the scores, and selects the results with the top ranking and sends them to the result presentation unit;

The result presentation unit, which presents different forms according to the types of results sent by the sorting unit.

Further, the sorting unit includes a questioning sub-unit, which is used to calculate whether there is a difference in the attribute value of each node or the attribute value of a hop associated entity in the recalled knowledge graph entity node set; if there is a difference, there will be a difference in the attribute value. As the content of the question, it is pushed to the user, and according to the user's answer to the question, the corresponding answer is screened and presented to the next step.

Further, the presentation type of the result presentation unit includes text, picture, video, and audio form.

A design method of an intelligent question answering robot in a car maintenance scene based on a knowledge graph, comprising the following steps:

S1. Build a car maintenance knowledge map:

S11. Ontology design: Design the auto maintenance knowledge map ontology based on the scene characteristics of the auto maintenance field;

S12. Knowledge extraction: extract node attributes and node relationship information in the knowledge graph from structured, unstructured and semi-structured maintenance data;

S13. Knowledge storage: store the extracted data in the graph database to form the final usable auto maintenance knowledge graph;

S2. Intelligent question-answering robot development: develop intelligent robots with the constructed auto maintenance knowledge graph as the knowledge base.

Further, in the ontology design of step S11 , which concepts exist in the knowledge graph and the existing relationships between these concepts are described through ontology.

Further, in the knowledge extraction in the step S12, if the data for saving knowledge is structured data, a template is written, and triples are extracted from it; if the data for saving knowledge is unstructured data, then a customized programming language is used to write and parse. Scripts, parsing raw data, structuring it.

Further, in the knowledge extraction in step S12, if the data storing the knowledge is unstructured data, entities and relationships are extracted from the unstructured data by training entity extraction and relationship extraction algorithm models.

Further, in the knowledge storage in step S13, the data obtained in the knowledge extraction step is converted into an input format supported by the graph database, and stored in the graph database.

Further, the step S2 intelligent question answering robot development includes:

S21. Natural language parsing: The robot conversation engine first understands the user's input intent, and converts the single or multiple rounds of text data input by the user into one or more graph query statements and corresponding parameters;

S22. Recall: input the graph query statement generated in the previous step into the knowledge graph, and receive the returned results of the query, and perform preliminary screening of the results;

S23. Sorting: For the recalled results, use the semantic engine to calculate and score the similarity, sort according to the scores, and select the results with the highest ranking;

S24. Answer presentation: different forms of presentation are performed according to the type of the returned result.

Further, in the natural language parsing of step S21, the robot conversation engine understands the user's intention through an entity linking algorithm.

Further, the step S23 sorting is provided with a questioning function: in the set of knowledge graph entity nodes of the recalled knowledge graph, whether there is a difference in the attribute value of each node or the attribute value of a hop associated entity; if there is a difference, there will be a difference. Attributes are used as the content of the question, and the user is asked, and according to the user's answer to the question, the corresponding answer is screened and presented to the next step.

Example 1

An intelligent question answering robot based on knowledge graph for automobile maintenance scene, including:

The maintenance knowledge graph module, which combines the scene characteristics of the maintenance field, establishes the knowledge graph ontology, and extracts the node attributes and node-related information in the knowledge graph ontology and stores it in the graph database.

The maintenance knowledge map module includes:

The knowledge graph ontology unit, which includes the failure phenomena, fault codes, parts, maintenance cases, vehicle series and other nodes of automobile maintenance, and includes the node relationship between each node.

Ontology is the schema of knowledge graph. In theory, ontology refers to "a formalized, explicit and detailed specification of a shared conceptual system". In layman's terms, an ontology describes which concepts exist in a knowledge graph, what relationships exist between these concepts, and so on. In this embodiment, as shown in Figure 2, the main body mainly includes concepts such as "fault phenomenon", "fault code", "parts", "maintenance case", "vehicle series"; among which, "maintenance case" also includes : "Failure Frequency", "Check Item", "Maintenance Plan" and other attributes.

A knowledge extraction unit, which extracts relevant information such as node attributes and node relationships from the knowledge graph ontology unit.

The knowledge extraction unit performs knowledge extraction from structured, unstructured and semi-structured data. In this embodiment, from the "maintenance case", the "fault code" related to the "fault phenomenon" is extracted to form a triplet of ["fault phenomenon", "related fault code", "fault code"].

If the data stored in the knowledge graph ontology unit is structured such as excel and json, a template can be written to extract triples from it; if the data stored in knowledge is unstructured data, it can be customized by programming languages such as python. Parse the script, parse the original data through regular expressions, rule functions, etc., and structure it. Further, it can also label data, train entity extraction and relationship extraction algorithm models, and extract entities and relationships from unstructured data. The entity extraction algorithm models include but are not limited to: HMM, LSTM-CRF, etc.; the relationship extraction algorithm models include but are not limited to: CR-CNN, Attention CNNs, etc.

The graph database unit is used for the knowledge information extracted by the knowledge extraction unit to form the final maintenance knowledge graph.

In this embodiment, the triples obtained in the knowledge extraction step are written in a programming language such as python, converted into an input format supported by the graph database, and stored in the graph database. A graph database is a database for semantic querying using a graph structure, which uses nodes, edges, and attributes to represent and store data. The key concept of this system is the graph, which directly associates data items in storage with data nodes and sets of edges that represent relationships between nodes. These relationships allow data in the store to be directly linked together and, in many cases, retrieved in one operation. The graph generated in this step is the maintenance knowledge graph, which can be used as the knowledge base of the intelligent question answering robot.

an intelligent question answering module, which is connected in communication with the maintenance knowledge graph module, performs language analysis on the input voice or text, and performs query retrieval from the maintenance knowledge graph module, and sorts and outputs the query retrieval results;

As shown in Figure 3, the intelligent question answering module includes:

A natural language parsing unit, which converts single or multiple rounds of consultation content (text data) input by the user into one or more graph query statements (including but not limited to: gremlin) and corresponding parameters.

Understanding user intent is achieved through entity linking algorithms. Entity linking refers to analyzing the text input by the user, identifying the entities in it, and returning the information of the associated entity nodes in the graph. By judging whether the entity type mentioned in the user's input is "fault phenomenon", "fault code" or "component", the user's intention can be judged. For example, if the user inputs "high engine water temperature", and the entity link returns an entity whose type is "fault phenomenon" and whose name is "high engine water temperature" in the map, it is judged that the user's intention is to consult the fault phenomenon. After clarifying the user's intent, a graph query statement is generated for the next recall.

The recall unit inputs the graph query statement generated by the natural language parsing unit into the maintenance knowledge graph module, receives the returned result of the query, and performs preliminary screening of the result. For example, executing a graph query statement will return failure type entities with names such as "high engine water temperature", "high vehicle water temperature", and "high engine temperature", as well as their associated maintenance case entities in one hop and parts entities associated with them in two hops. Take these entities as answer candidates.

A sorting unit, which performs similarity calculation and scoring with the help of the semantic engine of the results after the preliminary screening of the recalling unit, sorts the results according to the scores, selects the top-ranked results and returns them to the business front-end.

The sorting unit is designed with a questioning sub-unit, which is used to calculate whether there is a difference in the attribute value of each node or the attribute value of a hop associated entity in the recalled knowledge graph entity node set. If there is, the user will be questioned by using the attribute with the difference as the question content. And according to the user's answer to the question, the corresponding answer is screened and presented to the next step.

The result presentation unit, the business front-end presents different forms according to the type of the returned result. Presentation types include text, pictures, video, audio and other forms.

The intelligent Q&A module can conduct multiple rounds of Q&A interaction with customers to obtain solutions and related cases that can be referenced for fault maintenance.

Example 2

As shown in Figure 1, a design method of an intelligent question-answering robot in a car maintenance scene based on a knowledge graph includes the following steps:

S1. Build a car maintenance knowledge map:

S11. Ontology design: Design the auto maintenance knowledge map ontology based on the scene characteristics of the auto maintenance field;

S12. Knowledge extraction: extract node attributes and node relationship information in the knowledge graph from structured, unstructured and semi-structured maintenance data;

S13. Knowledge storage: store the extracted data in the graph database to form the final usable auto maintenance knowledge graph;

S2. Intelligent question-answering robot development: develop intelligent robots with the constructed auto maintenance knowledge graph as the knowledge base.

In the ontology design of step S11 , which concepts exist in the knowledge graph and the existing relationships between these concepts are described through ontology.

The ontology includes concepts such as "fault phenomenon", "fault code", "parts", "maintenance case", and "vehicle series".

In the knowledge extraction in step S12, if the data for saving knowledge is structured such as excel, json, etc., a template can be written, and triples can be extracted from it; Customized writing parsing scripts, parsing raw data through regular expressions, rule functions, etc., and structuring it.

Further, it can also label data, train entity extraction and relationship extraction algorithm models, and extract entities and relationships from unstructured data. Among them, entity extraction algorithm models include but are not limited to: HMM, LSTM-CRF, etc.; relation extraction algorithm models include but are not limited to: CR-CNN, Attention CNNs, etc.

In the knowledge storage in step S13, the triples obtained in the knowledge extraction step are written in a programming language such as python, converted into an input format supported by the graph database, and stored in the graph database.

A graph database is a database for semantic querying using a graph structure, which uses nodes, edges, and attributes to represent and store data. The key concept of this system is the graph, which directly associates data items in storage with data nodes and sets of edges that represent relationships between nodes. These relationships allow data in the store to be directly linked together and, in many cases, retrieved in one operation. The graph generated in this step is the maintenance knowledge graph, which can be used as the knowledge base of the intelligent question answering robot.

The step S2 of developing an intelligent question answering robot includes:

S21. Natural language parsing: The robot conversation engine will first understand the intent input by the user, and convert the single or multi-round consultation content (text data) input by the user into one or more graph query statements (including but not limited to: gremlin ) and the corresponding parameters. Among them, understanding user intent is achieved through entity linking algorithm. Entity linking refers to analyzing the text input by the user, identifying the entities in it, and returning the information of the associated entity nodes in the graph. By judging whether the entity type mentioned in the user's input is "fault phenomenon", "fault code" or "component", the user's intention can be judged. For example, if the user inputs "high engine water temperature", and the entity link returns an entity whose type is "fault phenomenon" and whose name is "high engine water temperature" in the map, it is judged that the user's intention is to consult the fault phenomenon. After clarifying the user's intent, a graph query statement is generated for the next recall.

S22. Recall: Input the graph query statement generated in the previous step into the knowledge graph, and receive the returned results of the query, and conduct preliminary screening of the results. For example, executing a graph query statement will return failure type entities with names such as "high engine water temperature", "high vehicle water temperature", and "high engine temperature", as well as their associated maintenance case entities in one hop and parts entities associated with them in two hops. Take these entities as answer candidates.

S23. Sorting: For the recalled results, use the semantic engine to calculate and score the similarity, sort according to the scores, select the top results and return them to the business front-end. At this stage, a questioning module is designed to calculate whether there is a difference in the attribute value of each node or the attribute value of a hop associated entity in the set of knowledge graph entity nodes to be recalled. If there is, the user will be questioned by using the attribute with the difference as the question content. And according to the user's answer to the question, the corresponding answer is screened and presented to the next step.

S24. Answer presentation: The business front-end presents different forms of results according to the type of the returned result. Presentation types include text, pictures, video, audio and other forms.

The question-and-answer robot can intelligently understand the intention of the user's input, perform the corresponding operation, query the answer from the knowledge base and return it to the user, helping the front-line maintenance technician to quickly obtain the repair method and related information of the fault.

Although embodiments of the present invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, and substitutions can be made in these embodiments without departing from the principle and spirit of the invention and modifications, the scope of the present invention is defined by the appended claims and their equivalents.

Claims (10)

1. The utility model provides a car maintenance scene intelligence question-answering robot based on knowledge map which characterized in that includes:

the maintenance knowledge graph module is used for establishing a knowledge graph body by combining the scene characteristics of the maintenance field, extracting the node attributes and the node related information in the knowledge graph body and storing the node attributes and the node related information into a graph database;

and the intelligent question-answering module is in communication connection with the maintenance knowledge map module, performs language analysis on input voice or characters, performs query retrieval from the maintenance knowledge map module, and outputs query retrieval results after sequencing.

2. The wisdom-based auto repair scenario intelligent question-answering robot of claim 1, wherein the repair wisdom module comprises:

the knowledge graph body unit comprises nodes for automobile maintenance and node relations among the nodes; the node comprises: fault phenomena, fault codes, parts, maintenance cases, vehicle systems;

the knowledge extraction unit extracts relevant information such as node attributes and node relations from the knowledge graph body unit;

and the database unit is used for forming a final maintenance knowledge map by the knowledge information extracted by the knowledge extraction unit.

3. The intellectual property map based automobile maintenance scene question-answering robot as claimed in claim 2, wherein the map database is a database semantically queried using a graph structure, which uses nodes, edges and attributes to represent and store data.

4. The knowledge-graph-based automotive maintenance scenario intelligent question-answering robot of claim 1, wherein the intelligent question-answering module comprises:

the natural language analysis unit converts single-round or multi-round text data input by a user into one or more graph query sentences and corresponding parameters;

the recall unit inputs the graph query sentence generated by the natural language analysis unit into the maintenance knowledge graph module for retrieval and query, receives a return result of the query and performs result primary screening;

the sorting unit is used for calculating the similarity and scoring the results preliminarily screened by the recall unit, sorting according to the scores, selecting the results sorted in the front and sending the results to the result presentation unit;

and the result presentation unit is used for presenting different forms according to the types of the results sent by the sorting unit.

5. The intelligent knowledge-graph-based question-answering robot for automobile maintenance scenes as claimed in claim 4, wherein the sorting unit comprises a question-pursuing subunit for calculating whether the attribute values of the nodes or the attribute values of the associated entities in the recalled knowledge-graph entity node sets are different; if the difference exists, the attribute with the difference is used as the question hunting content and pushed to the user, and the corresponding answer is screened for the next step to be presented according to the answer of the user to the question hunting.

6. The design method of the knowledge-graph-based intelligent question-answering robot for the automobile maintenance scene as claimed in claim 1, characterized by comprising the following steps:

s1, constructing an automobile maintenance knowledge map:

s11, designing a body: designing an automobile maintenance knowledge map body by combining scene characteristics of the automobile maintenance field;

s12, knowledge extraction: extracting node attributes and node relation information in the knowledge graph from structured, unstructured and semi-structured maintenance data;

s13, knowledge storage: storing the extracted data into a database to form a final usable automobile maintenance knowledge map;

s2, intelligent question and answer robot development: and developing the intelligent robot by taking the constructed automobile maintenance knowledge map as a knowledge base.

7. The method as claimed in claim 6, wherein in the step S12 of extracting knowledge,

if the data holding knowledge is structured data: compiling a template and extracting triples from the template;

if the data holding knowledge is unstructured data: compiling an analysis script through programming language customization, analyzing the original data and structuring the original data; or extracting the entities and the relations from the unstructured data by training an entity extraction and relation extraction algorithm model.

8. The method as claimed in claim 6, wherein in the step S13, the data obtained in the step of knowledge extraction is converted into an input format supported by a database and stored in the database.

9. The design method of the knowledge-graph-based intelligent question-answering robot for the automobile maintenance scene as claimed in claim 6, wherein the step S2 development of the intelligent question-answering robot comprises:

s21, natural language analysis: the robot conversation engine firstly understands the intention input by a user and converts single-round or multi-round text data input by the user into one or more graph query sentences and corresponding parameters;

s22, recalling: inputting the graph query statement generated in the last step into a knowledge graph, receiving a returned result of query, and performing result primary screening;

s23, sequencing: for the recalled results, similarity calculation and scoring are carried out by means of a semantic engine, the results are sorted according to the scores, and the results sorted in the front are selected;

s24, answer presentation: and according to the type of the returned result, different forms of presentation are performed.

10. The method for designing the intelligent questioning and answering robot for automobile maintenance scene based on knowledge graph according to claim 9, wherein the step S23 is sequenced and provided with the following question function: calculating whether the attribute values of all nodes or the attribute values of one-hop associated entities in the recalled knowledge graph entity node set have differences or not; if the difference exists, the attribute with the difference is used as the question hunting content to hunt for the user, and the corresponding answer is screened for the next step to be presented according to the answer of the user to the question hunting.

Images