Large model knowledge question and answer optimization method based on semantic enhanced knowledge graph

Through the knowledge graph method based on semantic enhancement, the problems of low accuracy and insufficient integrity of knowledge graph data vectorization in the existing technology are solved, and a more efficient knowledge question-and-answer system is realized. Through the combination of data segmentation and multiple search methods, the The data quality of the knowledge base and the accuracy of the search results.

CN118939779AInactive Publication Date: 2024-11-12BEIJING DETA JINGYAO INFORMATION TECH CO LTD

Patent Information

Application Number
CN202411236334.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-04
Publication Date
2024-11-12
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

In the data vectorization method, the existing knowledge graph has low accuracy, easy to be missing, and it is difficult to effectively display the dynamic development laws of the knowledge field.

Method used

Using a knowledge graph method based on semantic enhancement, we use the enterprise data API interface data to obtain enterprise data, segment and save it to vector database, index database and knowledge graph database, and combine question enhancement and multiple search methods to generate integrated answers using big models.

Benefits of technology

Improve the accuracy and completeness of the knowledge base data quality and search results, providing more accurate and comprehensive knowledge Q&A.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118939779A_ABST
    Figure CN118939779A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of knowledge graph construction, in particular to a large model knowledge question and answer optimization method based on a semantic enhanced knowledge graph, and the optimization method comprises the steps: obtaining data of an enterprise data API interface; based on API interface data, segmenting the data, and respectively storing the data in a vector database, an index database and a knowledge graph database; the user puts forward questions, and the questions put forward by the user are enhanced; performing vector retrieval, index retrieval and graph retrieval based on the enhanced problem; and rearranging and formatting the combined data, and performing integrated answering through a large model. According to the method, the knowledge graph database is constructed, comprehensive detection including vector retrieval, index retrieval and graph retrieval can be carried out on questions, the data quality of the knowledge base can be improved, in addition, in the retrieval stage, the accuracy of the retrieval result and the answer integrity can be improved, and better commercial prospects are brought.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of knowledge graph construction, and specifically is a large-model knowledge question-answering optimization method based on semantically enhanced knowledge graph. Background Art

[0002] Knowledge graph, known as knowledge domain visualization or knowledge domain mapping map in the library and information community, is a series of various graphics that show the development process and structural relationship of knowledge. It uses visualization technology to describe knowledge resources and their carriers, and to mine, analyze, construct, draw and display knowledge and their interrelationships. Specifically, knowledge graph is a modern theory that combines the theories and methods of applied mathematics, graphics, information visualization technology, information science and other disciplines with metrological citation analysis, co-occurrence analysis and other methods, and uses visualized graphs to vividly display the core structure, development history, frontier fields and overall knowledge architecture of the discipline to achieve the purpose of multidisciplinary integration. It displays complex knowledge fields through data mining, information processing, knowledge metrology and graphics drawing, reveals the dynamic development laws of knowledge fields, and provides practical and valuable references for disciplinary research. So far, its practical application has been gradually expanded in developed countries and has achieved good results, but it is still in the initial stage of research in my country.

[0003] Currently, most methods use data vectorization, which uses vector search to compare questions and knowledge base data. However, the accuracy is not high and the integrity is easily lost. Summary of the invention

[0004] In view of the shortcomings of the prior art, the present invention provides a large-model knowledge question-answering optimization method based on a semantically enhanced knowledge graph to solve the above technical problems.

[0005] To achieve the above-mentioned purpose, the present invention provides the following technical solutions: a large-model knowledge question-answering optimization method based on semantically enhanced knowledge graph, the optimization method comprising: S1: obtaining data from the enterprise data API interface; S2: based on the API interface data, segmenting the data, and saving them to a vector database, an index database, and a knowledge graph database respectively; S3: when a user asks a question, the question raised by the user is enhanced; S4: based on the enhanced question, vector retrieval, index retrieval, and graph retrieval are performed; S5: the combined data is rearranged and formatted, and an integrated answer is given through a large model.

[0006] Preferably, the step of S1 is: Register and log in to the developer website of the service provider that provides the enterprise data API; Create a new application or project and obtain the keys needed to access the API; Consult the API documentation for information on endpoints, request parameters, response formats, and limitations; Use HTTP or HTTPS protocol to construct requests that comply with the requirements of the API documentation; Use the HTTP library to send a constructed request to the API endpoint; Receive responses from the API and return data in JSON format; Parse the API response and extract the data.

[0007] Preferably, the step of S2 is: Clean and format the data obtained from the API; According to the determined basis, the data segmentation operation is performed. The segmentation here is not based on the time period, but on the meaning of the text. The segmentation methods include: fixed character length, separator method, and fixed character length plus repeated characters before and after segmentation; Save to vector database; Extract the stored vector data from the data obtained from the API; Use the API or client library provided by the database to insert vector data into the database; According to the application scenario and query requirements, set the index to optimize the retrieval performance of the vector; Design the database table structure and define fields and data types based on the data characteristics obtained from the API; Insert the data obtained from the API into the corresponding database table and save it in the database; Import the data obtained from the API into the knowledge graph database to store and query the relationship data between entities.

[0008] Preferably, the data obtained from the API is imported into the knowledge graph database, and the relationship data between entities to be stored and queried includes: Identification analysis, analyzing and identifying entities and relationships from enterprise data; Graph data processing requires deduplication and synonym recognition, and finally storage in the graph database; Identification analysis includes: Entity recognition: First, the enterprise data is divided into blocks. According to the token number limit of the large model base, the blocks should be as large as possible. Fine-tuned instructions are used to combine the block data and then handed over to the large model for entity recognition. Entity prediction: Given a start entity and a relation, the task is to predict the end entity related to the start entity through a large model; Relationship identification: Slice the enterprise data, use fine-tuned instructions to combine the sliced ​​data, and hand it over to the big model for relationship identification; Relationship prediction: Given a start entity and an end entity, the task is to predict the relationship between them through a large model; Triplet verification: given a triplet (h, r, t), classify it as correct or incorrect through the large model; Graph data processing includes: Vector representation: Using the vector model, the retrieval processor converts each entity word and relationship name into a vector representation and then stores it in the graph database; Similarity matching: The retrieval processor uses cosine similarity to calculate the similarity between the query term and the nodes in the knowledge graph. The higher the similarity, the more relevant the query term and the node are, and node and relationship merging is performed. The processed data is saved in a graph database. All nodes have an embedding attribute that stores the vector representation of the node information. The graph database uses NEO4J.

[0009] Preferably, the vector model formula is: For entity words and relationship name , their vectors are represented as and Generated by the Word2Vec model respectively; Entity words Nodes stored as graph databases: Name the relationship Nodes stored as graph databases: Need to store relationship , the relationship can be Stored as edges in a graph database, where Can be used as one of the attributes of an edge.

[0010] Preferably, the cosine similarity calculation includes: The formula for calculating cosine similarity is:

[0011] in, Represents the query vector and node vector The dot product (inner product) of and are query vectors respectively. and node vector The norm of is the cosine similarity, and the value range of cosine similarity is between -1 and 1. The closer it is to 1, the more similar they are, and the closer it is to -1, the less similar they are. The higher the similarity, the more relevant the query term and the node are.

[0012] Preferably, the step of S4 is: Convert the augmented problem into a vector representation; The documents or entities are sorted based on the similarity scores calculated by cosine similarity, and a threshold is set to filter out items with low similarity; First, you need to create an index for documents or entities, use the enhanced questions as queries, and quickly locate documents or entities containing relevant information through the index data structure; Search the knowledge graph to find entities or relationships related to the question; Represent entities and relations in the knowledge graph as a combination of nodes and edges, which are represented by vectors to capture their semantic information; The retrieved entities are filtered based on specific metrics of the query results and the most relevant results are returned to the user.

[0013] Preferably, graph retrieval includes: When a new question is received, the language model is used to understand the key points of the question, and the retrieval processor is used to associate these key points with the nodes in the knowledge graph. Finally, through the personalized algorithm, the information across multiple text fragments is quickly found and integrated in the knowledge graph to provide an accurate answer for the output; Graph retrieval processing is used for two main tasks: Extract entities from a query: Extract named entities from queries using a large language model fine-tuned with instructions; Associate entities with knowledge graph nodes: Use the pre-trained graph searcher to calculate the similarity between the PageRank algorithm query entity and the nodes in the knowledge graph, vectorize the entity information, access NEO4J through the CYPHER language, query the graph nodes that are similar to the entity information vector and the knowledge graph node embedding attribute vector, and obtain the TopN nodes with the highest similarity and the relationship between them. The search processor captures the subtle differences between entities, thereby associating the query entity with the related nodes in the knowledge graph. To run a custom PageRank algorithm: Use the query entity as the seed node and run a custom PageRank algorithm on the TopN results. The PageRank algorithm calculates the probability value of each node based on the association between nodes, thereby identifying the nodes most relevant to the query entity. The custom PageRank algorithm gives priority to nodes related to the query entity, thereby improving the retrieval quality.

[0014] Preferably, the PageRank algorithm includes: The PageRank algorithm formula is:

[0015] in Is a node The PageRank value of is pointing to The node set of Is a node How many links are there? is the damping factor, set to 0.85, is the total number of nodes in the graph, is a relative relation variable, where all elements are ; For the query entity and the nodes directly connected to the query entity, set to ,According to different data sets, use appropriate setting values ​​through data testing; A set of entities and relationships are obtained through the graph retrieval processor. A pre-trained re-ranking model is used to prioritize the selected entity and relationship information, and the TopN with the highest priority are selected. Then, the selected entities and relationships are integrated into a text to provide an information basis for large model queries.

[0016] Preferably, the formatted data is input into a large model, and the large model is used to generate integrated answers, which includes understanding the input questions and generating answers, and post-processing the answers, including formatting output, answer sorting and filtering.

[0017] Compared with the prior art, the present invention has the following beneficial effects: This application can improve the data quality of the knowledge base itself by constructing a knowledge graph database. In addition, it can improve the accuracy of retrieval results and the completeness of answers during the retrieval stage. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a step flow framework diagram of the present invention. DETAILED DESCRIPTION

[0019] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0020] Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in the field without making any creative work shall fall within the scope of protection of the present invention.

[0021] The present invention provides a technical solution: a large model knowledge question answering optimization method based on semantically enhanced knowledge graph, the optimization method comprising: S1: Get data from the enterprise data API interface; S2: Based on the API interface data, the data is segmented and saved to the vector database, index database and knowledge graph database respectively; S3: Users raise questions, and the questions raised by users are enhanced; S4: Based on the enhanced questions, perform vector retrieval, index retrieval, and graph retrieval; S5: Rearrange and format the combined data, and make an integrated answer through a large model.

[0022] Furthermore, the steps of S1 are: Register and log in to the developer website of the service provider that provides the enterprise data API; Create a new application or project and obtain the keys needed to access the API; Consult the API documentation for information on endpoints, request parameters, response formats, and limitations; Use HTTP or HTTPS protocol to construct requests that comply with the requirements of the API documentation; Use the HTTP library to send a constructed request to the API endpoint; Receive responses from the API and return data in JSON format; Parse the API response and extract the data.

[0023] Furthermore, the steps of S2 are: Clean and format the data obtained from the API; According to the determined basis, the data segmentation operation is performed. The segmentation here is not based on the time period, but on the meaning of the text. The segmentation methods include: fixed character length, separator method, and fixed character length plus repeated characters before and after segmentation; Save to vector database; Extract the stored vector data from the data obtained from the API; Use the API or client library provided by the database to insert vector data into the database; According to the application scenario and query requirements, set the index to optimize the retrieval performance of the vector; Design the database table structure and define fields and data types based on the data characteristics obtained from the API; Insert the data obtained from the API into the corresponding database table and save it in the database; Import the data obtained from the API into the knowledge graph database to store and query the relationship data between entities.

[0024] Furthermore, the data obtained from the API is imported into the knowledge graph database to store and query the relationship data between entities, including: Identification analysis, analyzing and identifying entities and relationships from enterprise data; Graph data processing requires deduplication and synonym recognition, and finally storage in the graph database; Identification analysis includes: Entity recognition: First, the enterprise data is divided into blocks. According to the token number limit of the large model base, the blocks should be as large as possible. Fine-tuned instructions are used to combine the block data and then handed over to the large model for entity recognition. Entity prediction: Given a start entity and a relation, the task is to predict the end entity related to the start entity through a large model; Relationship identification: Slice the enterprise data, use fine-tuned instructions to combine the sliced ​​data, and hand it over to the big model for relationship identification; Relationship prediction: Given a start entity and an end entity, the task is to predict the relationship between them through a large model; Triplet verification: given a triplet (h, r, t), classify it as correct or incorrect through the large model; Graph data processing includes: Vector representation: Using the vector model, the retrieval processor converts each entity word and relationship name into a vector representation and then stores it in the graph database; Similarity matching: The retrieval processor uses cosine similarity to calculate the similarity between the query term and the nodes in the knowledge graph. The higher the similarity, the more relevant the query term and the node are, and node and relationship merging is performed. The processed data is saved in a graph database. All nodes have an embedding attribute that stores the vector representation of the node information. The graph database uses NEO4J.

[0025] Furthermore, the vector model formula is: For entity words and relationship name , their vectors are represented as and Generated by the Word2Vec model respectively; Entity words Nodes stored as graph databases: Name the relationship Nodes stored as graph databases: Need to store relationship , the relationship can be Stored as edges in a graph database, where Can be used as one of the attributes of an edge.

[0026] Furthermore, the cosine similarity calculation includes: The formula for calculating cosine similarity is:

[0027] in, Represents the query vector and node vector The dot product (inner product) of and are query vectors respectively. and node vector The norm of is the cosine similarity, and the value range of cosine similarity is between -1 and 1. The closer it is to 1, the more similar they are, and the closer it is to -1, the less similar they are. The higher the similarity, the more relevant the query term and the node are.

[0028] Furthermore, the steps of S4 are: Convert the augmented problem into a vector representation; The documents or entities are sorted based on the similarity scores calculated by cosine similarity, and a threshold is set to filter out items with low similarity; First, you need to create an index for documents or entities, use the enhanced questions as queries, and quickly locate documents or entities containing relevant information through the index data structure; Search the knowledge graph to find entities or relationships related to the question; Represent entities and relations in the knowledge graph as a combination of nodes and edges, which are represented by vectors to capture their semantic information; The retrieved entities are filtered based on specific metrics of the query results and the most relevant results are returned to the user.

[0029] Furthermore, graph retrieval includes: When a new question is received, the language model is used to understand the key points of the question, and the retrieval processor is used to associate these key points with the nodes in the knowledge graph. Finally, through the personalized algorithm, the information across multiple text fragments is quickly found and integrated in the knowledge graph to provide an accurate answer for the output; Graph retrieval processing is used for two main tasks: Extract entities from a query: Extract named entities from queries using a large language model fine-tuned with instructions; Associate entities with knowledge graph nodes: Use the pre-trained graph searcher to calculate the similarity between the PageRank algorithm query entity and the nodes in the knowledge graph, vectorize the entity information, access NEO4J through the CYPHER language, query the graph nodes that are similar to the entity information vector and the knowledge graph node embedding attribute vector, and obtain the TopN nodes with the highest similarity and the relationship between them. The search processor captures the subtle differences between entities, thereby associating the query entity with the related nodes in the knowledge graph. To run a custom PageRank algorithm: Use the query entity as the seed node and run a custom PageRank algorithm on the TopN results. The PageRank algorithm calculates the probability value of each node based on the association between nodes, thereby identifying the nodes most relevant to the query entity. The custom PageRank algorithm gives priority to nodes related to the query entity, thereby improving the retrieval quality.

[0030] Furthermore, the PageRank algorithm includes: The PageRank algorithm formula is:

[0031] in Is a node The PageRank value of is pointing to The node set of Is a node How many links are there? is the damping factor, set to 0.85, is the total number of nodes in the graph, is a relative relation variable, where all elements are ; For the query entity and the nodes directly connected to the query entity, set to ,According to different data sets, use appropriate setting values ​​through data testing; A set of entities and relationships are obtained through the graph retrieval processor. A pre-trained re-ranking model is used to prioritize the selected entity and relationship information, and the TopN with the highest priority are selected. Then, the selected entities and relationships are integrated into a text to provide an information basis for large model queries.

[0032] Furthermore, the formatted data is input into the big model, and the big model is used to generate integrated answers, which includes understanding the input questions, generating answers, and post-processing the answers, including formatting output, answer sorting, and filtering.

[0033] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0034] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A large-model knowledge question answering optimization method based on semantically enhanced knowledge graph, characterized in that: Optimization methods include: S1: Get data from the enterprise data API interface; S2: Based on the API interface data, the data is segmented and saved to the vector database, index database and knowledge graph database respectively; S3: Users raise questions, and the questions raised by users are enhanced; S4: Based on the enhanced questions, perform vector retrieval, index retrieval, and graph retrieval; S5: Rearrange and format the combined data, and make an integrated answer through a large model.

2. The large-model knowledge question answering optimization method based on semantically enhanced knowledge graph according to claim 1 is characterized in that: The steps of S1 are: Register and log in to the developer website of the service provider that provides the enterprise data API; Create a new application or project and obtain the keys needed to access the API; Consult the API documentation for information on endpoints, request parameters, response formats, and limitations; Use HTTP to construct requests that comply with the API documentation requirements; Use the HTTP library to send a constructed request to the API endpoint; Receive responses from the API and return data in JSON format; Parse the API response and extract the data.

3. The large-model knowledge question answering optimization method based on semantically enhanced knowledge graph according to claim 1 is characterized in that: The steps of S2 are: Clean and format the data obtained from the API; According to the determined basis, a data segmentation operation is performed. According to the determined basis, a data segmentation operation is performed; Save to vector database; Extract the stored vector data from the data obtained from the API; Use the API provided by the database to insert vector data into the database; According to the application scenario and query requirements, set the index to optimize the retrieval performance of the vector; Design the database table structure and define fields and data types based on the data characteristics obtained from the API; Insert the data obtained from the API into the corresponding database table and save it in the database; Import the data obtained from the API into the knowledge graph database to store and query the relationship data between entities.

4. The large-model knowledge question answering optimization method based on semantically enhanced knowledge graph according to claim 3 is characterized in that: Import the data obtained from the API into the knowledge graph database to store and query the relationship data between entities including: Identification analysis, analyzing and identifying entities and relationships from enterprise data; Graph data processing requires deduplication and synonym recognition, and finally storage in the graph database; Identification analysis includes: Entity recognition: First, the enterprise data is divided into blocks. According to the token number limit of the large model base, the blocks need to be enlarged. Fine-tuned instructions are used to combine the block data and then handed over to the large model for entity recognition. Entity prediction: predict the end entity related to the start entity through the large model; Relationship identification: Slice the enterprise data, use fine-tuned instructions to combine the sliced ​​data, and hand it over to the big model for relationship identification; Relationship prediction: Given a start entity and an end entity, predict the relationship between them through a large model; Triplet verification: given a triplet (h, r, t), classify it as correct or incorrect through the large model; Graph data processing includes: Vector representation: Using the vector model, the retrieval processor converts each entity word and relationship name into a vector representation and then stores it in the graph database; Similarity matching: The retrieval processor uses cosine similarity to calculate the similarity between the query term and the nodes in the knowledge graph. The higher the similarity, the more relevant the query term and the node are, and node and relationship merging is performed; The processed data is saved in a graph database. All nodes have an embedding attribute that stores the vector representation of the node information. The graph database uses NEO4J.

5. The large-model knowledge question answering optimization method based on semantically enhanced knowledge graph according to claim 4 is characterized in that: The vector model formula is: For entity words and relationship name , their vectors are represented as and Generated by the Word2Vec model respectively; Entity words Nodes stored as graph databases: Name the relationship Nodes stored as graph databases: Need to store relationship , the relationship can be Stored as edges in a graph database, where Can be used as one of the attributes of an edge.

6. The large-model knowledge question answering optimization based on semantically enhanced knowledge graph according to claim 4 The method is characterized in that The cosine similarity calculation includes: The formula for calculating cosine similarity is: in, Represents the query vector and node vector The dot product (inner product) of and are query vectors respectively. and node vector The norm of is the cosine similarity, and the value range of cosine similarity is between -1 and 1. The closer it is to 1, the more similar they are, and the closer it is to -1, the less similar they are. The higher the similarity, the more relevant the query term and the node are.

7. The large-model knowledge question answering optimization method based on semantically enhanced knowledge graph according to claim 1 is characterized in that: The steps of S4 are: Convert the augmented problem into a vector representation; The documents or entities are sorted based on the similarity scores calculated by cosine similarity, and a threshold is set to filter out items with low similarity; First, you need to create an index for the document entity, use the enhanced question as a query, and quickly locate the document or entity containing relevant information through the index data structure; Search the knowledge graph to find entities or relationships related to the question; Represent entities and relations in the knowledge graph as a combination of nodes and edges, which are represented by vectors to capture their semantic information; Filter the retrieved entities according to specific metrics of the query results.

8. The large-model knowledge question answering optimization method based on semantically enhanced knowledge graph according to claim 4 is characterized in that: Graph retrieval includes: When a new question is received, the language model is used to understand the key points of the question, and the retrieval processor is used to associate these key points with the nodes in the knowledge graph. Finally, through the personalized algorithm, the information across multiple text fragments is quickly found and integrated in the knowledge graph to provide an accurate answer for the output; Graph retrieval processing is used for two main tasks: Extract entities from a query: Extract named entities from queries using a large language model fine-tuned with instructions; Associate entities with knowledge graph nodes: Use the pre-trained graph searcher to calculate the similarity between the PageRank algorithm query entity and the nodes in the knowledge graph, vectorize the entity information, access NEO4J through the CYPHER language, query the graph nodes that are similar to the entity information vector and the knowledge graph node embedding attribute vector, and obtain the TopN nodes with the highest similarity and the relationship between them. The search processor captures the subtle differences between entities, thereby associating the query entity with the related nodes in the knowledge graph. To run a custom PageRank algorithm: Use the query entity as the seed node and run a custom PageRank algorithm on the TopN results. The PageRank algorithm calculates the probability value of each node based on the association between nodes, thereby identifying the nodes most relevant to the query entity. The custom PageRank algorithm gives priority to nodes related to the query entity, thereby improving the retrieval quality.

9. The large-model knowledge question answering optimization method based on semantically enhanced knowledge graph according to claim 1 is characterized in that: The PageRank algorithm includes: The PageRank algorithm formula is: in Is a node The PageRank value of is pointing to The node set of Is a node How many links are there? is the damping factor, set to 0.85, is the total number of nodes in the graph, is a relative relation variable, where all elements are ; For the query entity and the nodes directly connected to the query entity, set to ,According to different data sets, use appropriate setting values ​​through data testing; A set of entities and relationships are obtained through the graph retrieval processor, and the selected entity and relationship information is prioritized, and the TopN with the highest priority are selected. Then, the selected entities and relationships are integrated into a text to provide an information basis for large model queries.

10. The large-model knowledge question answering optimization method based on semantically enhanced knowledge graph according to claim 1 is characterized by: Input the formatted data into the big model, and use the big model to generate integrated answers, which includes understanding the input questions and generating answers, and post-processing the answers, including formatting the output, sorting and filtering the answers.

Citation Information

Patent Citations

  • Steel industry intelligent question and answer method and device based on large model and medium

    CN118051587A

  • Geotechnical engineering knowledge question-answering system based on large language model

    CN118193704A

  • Intelligent customer service question and answer method based on knowledge graph and large language model

    CN118312600A

  • Knowledge graph question-answer method and apparatus based on deep learning technology, and device

    WO2021139283A1

Cited By

  • Knowledge graph and vector retrieval enhancement-based photovoltaic field large model efficiency improvement method

    CN119988639A

  • Enterprise data retrieval method and system based on large model and medium

    CN120067141A

  • An enterprise data retrieval method, system and medium based on large models

    CN120067141B

  • Decision-making auxiliary method and device based on large model

    CN120654795A

  • Knowledge question and answer high-correlation recommendation method based on large model semantic enhancement

    CN120873177A