Knowledge base knowledge association fusion method based on knowledge graph

Abstract

The invention discloses a knowledge base knowledge association fusion method based on a knowledge graph, and the method comprises the steps: integrating structured, semi-structured and non-structured data through a cross-modal alignment technology, and constructing a multi-source heterogeneous data association network of a unified semantic space; newly added external data are fused to a multi-source heterogeneous data association network in real time by using a dynamic attention mechanism, and entity conflicts are eliminated by combining rule reasoning and a machine learning model; hidden association among entities in the multi-source heterogeneous data association network is mined based on the graph neural network, and a knowledge graph logic chain is complemented; based on the knowledge graph, intelligent question and answer and risk assessment decision scenes are supported through a hybrid retrieval architecture and an inference engine; and automatically updating and associating the knowledge base of the network extension knowledge graph by adopting an incremental learning technology. Natural language questions and answers are supported, accurate answers are generated through knowledge reasoning, and user experience is remarkably enhanced.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of intelligent search and question answering systems, and particularly relates to a knowledge base knowledge association fusion method based on a knowledge graph. BACKGROUND

[0002] As a structured knowledge representation method, a knowledge graph can intuitively demonstrate the internal connection of knowledge by describing entities and their relationships through nodes and edges. A knowledge base is a system for storing and managing knowledge, providing a foundation for knowledge reasoning, querying, and application. By combining a knowledge graph with a knowledge base and using knowledge association fusion technology, the knowledge representation ability, query efficiency, and intelligence level of the knowledge base can be significantly improved.

[0003] In existing technologies, different knowledge sources have spelling, translation differences, and data integrity problems, making it difficult to completely resolve knowledge conflicts. For example, the same entity may be identified as multiple independent nodes due to different naming methods, requiring a large amount of manual or computational resources for alignment and disambiguation. In addition, there are a large number of redundant states in the fusion process, increasing the complexity of knowledge integration. For example, the same concept may be repeatedly expressed due to different sources, requiring complex rules or algorithms for deduplication and standardization. SUMMARY

[0004] The purpose of the present application is to improve the intelligence level of information retrieval, enhance the efficiency of knowledge recommendation and discovery, and optimize the knowledge update and maintenance process. The present application provides a knowledge base knowledge association fusion method based on a knowledge graph, which greatly improves the accuracy and relevance of search results through the semantic association ability of the knowledge graph.

[0005] The technical solution adopted by the present application is as follows: The knowledge base knowledge association fusion method based on a knowledge graph includes the following steps: Step 1: Integrate structured, semi-structured, and unstructured data through cross-modal alignment technology to construct a multi-source heterogeneous data association network in a unified semantic space; Specifically, through cross-modal alignment technology (such as the CLIP model), visual, linguistic, and audio data are mapped to a unified semantic space, eliminating text-image bias and audio-video description ambiguity, constructing a multi-source heterogeneous data association network, and supporting intelligent question answering accuracy improvement. Through knowledge abstraction modeling (RDF triples, attribute graphs) and embedding representation (low-dimensional vectors), cross-disciplinary and cross-language knowledge transfer is achieved, shortening the development cycle of multi-field applications.

[0006] Step 2: Use a dynamic attention mechanism to fuse time-series data (obtain new external data) into the multi-source heterogeneous data association network in real time, and combine rule-based reasoning and machine learning models to resolve entity conflicts; Specifically, a dynamic attention mechanism (such as the MCAN model) is used to fuse new knowledge into the knowledge graph in real time, and rule-based reasoning and machine learning are used to detect and resolve entity conflicts. Step 3: Based on graph neural network, the implicit association between entities in the multi-source heterogeneous data association network is mined, and the logical chain of the knowledge graph is completed. Specifically, based on graph neural network (GAT), multi-hop neighbor relationships are mined to discover implicit associations and complete missing logical chains in the knowledge graph, optimizing the accuracy of the recommendation system. Step 4: Based on the knowledge graph, a hybrid retrieval architecture (BM25 + vector + RAG) and reasoning engine are used to support intelligent question answering and risk assessment decision-making scenarios. Step 5: Incremental learning technology is used to automatically update and expand the knowledge base of the knowledge graph.

[0007] Further, step 1 specifically includes the following steps: Step 1-1: Obtain semi-structured and unstructured data for knowledge extraction to obtain knowledge extraction results; knowledge extraction includes entity extraction, relationship extraction between entities, and attribute extraction of entities. Step 1-2: Integrate the knowledge extraction results with structured data to obtain knowledge representation and eliminate information silos.

[0008] Step 1-3: Align entities and correct attributes of the knowledge representation to construct a multi-source heterogeneous data association network in a unified semantic space.

[0009] Further, the cross-modal alignment technology in step 1 uses the CLIP model to map visual, linguistic, and speech data to a unified semantic space, eliminating discrepancies between images and text descriptions and inconsistencies between audio and video.

[0010] Further, the heterogeneous data in step 1 includes visual data, linguistic data, and speech data, and the association network of multi-source heterogeneous data is constructed by first eliminating discrepancies between images and text and ambiguities in audio and video descriptions.

[0011] Further, in step 2, the dynamic attention mechanism uses the MCAN model to detect and fuse new knowledge in real time, automatically trigger the conflict resolution process, solve synonym entity naming conflicts or polysemy relationship ambiguities, and ensure the timeliness of the knowledge base.

[0012] Further, in step 3, the graph attention network (GAT) is used to analyze multi-hop neighbor relationships to generate knowledge completion suggestions and optimize the accuracy of the recommendation system.

[0013] Further, in step 4, the hybrid retrieval architecture specifically includes combining traditional retrieval (BM25), vector similarity calculation, and retrieval enhancement generation (RAG) model to improve the recall rate and accuracy of complex queries.

[0014] Further, the steps of the intelligent question answering in step 4 are as follows: Step 4-1, the server obtains the user's question and performs query analysis to extract keywords, and vectorizes the keywords of the user's question; Step 4-2, retrieve a plurality of recalled knowledge items from the vector database of the knowledge graph by the vectorized keywords; Step 4-3, reordering the plurality of recalled knowledge items according to the correlation degree from high to low; Step 4-4, generating a question answer according to the recalled knowledge items and returning it to the user. Specifically, the question answer is generated by a large language model to present a natural language answer to the user.

[0015] Further, the incremental learning in step 5 includes: developing an association network expansion algorithm to automatically discover new entities and new relationships, embedding and integrating the new knowledge into the knowledge graph through knowledge abstraction modeling (RDF triple, attribute graph) and embedding representation (low-dimensional vector).

[0016] The application adopts the above technical scheme, integrates multi-source heterogeneous data (such as documents, speech, images) and dynamically constructs a knowledge graph, uses cross-modal alignment technology and hybrid retrieval architecture to generate a knowledge graph, embeds the knowledge graph into an enterprise-level knowledge base reasoning process, constructs a dynamic semantic association network, supports knowledge fusion applications, and solves the problems of information silos and knowledge fragmentation. At the same time, the application integrates multi-modal data of different systems and formats, mines implicit associations based on graph neural networks, supports entity alignment, attribute correction, and semantic disambiguation, realizes continuous updating of knowledge assets and an intelligent decision-making closed loop, solves the problems of information silos, knowledge fragmentation, and updating lag, and supports intelligent decision-making and continuous value-added of knowledge assets in the scene.

[0017] The knowledge graph of the application supports natural language question answering, generates accurate answers through knowledge reasoning, and significantly enhances user experience. In the recommendation system, the technology uses the entity association and reasoning ability of the knowledge graph to realize personalized recommendation. BRIEF DESCRIPTION OF DRAWINGS

[0018] The application will be further described in detail below in combination with the drawings and specific embodiments; Figure 1 The principle architecture diagram of the knowledge base knowledge association fusion method based on the knowledge graph of the application; Figure 2 The function instance diagram of the knowledge base knowledge association fusion method based on the knowledge graph of the application. DETAILED DESCRIPTION

[0019] To make the purposes, technical solutions, and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application.

[0020] Traditional knowledge bases mainly rely on keyword matching, lack flexibility, and are difficult to cope with complex and variable natural language queries. Knowledge graphs represent knowledge through graphical means and visualize the relationships between knowledge points, providing structured support for information in knowledge bases. Users can not only visually see the associations between information through knowledge graphs and quickly locate the required knowledge, but also can reason through the relationships in the graph to obtain more accurate results. Knowledge recommendation in traditional knowledge bases is often based on simple statistical models or artificial rules, making it difficult to capture deep associations between knowledge nodes. Knowledge graphs can actively discover and recommend related knowledge content based on the associations between knowledge nodes. When a user searches for a certain topic in a knowledge base, based on the associations of the knowledge graph, the system can recommend other knowledge points related to the topic, thereby improving the efficiency of information discovery. The structured nature of knowledge graphs also makes content updating and maintenance in knowledge bases more efficient. Through the structure of the graph, enterprises can clearly understand the dependency relationships and impact range of different knowledge points, update and supplement the content in the knowledge base in a timely manner, and ensure that updates do not damage the overall consistency of the system.

[0021] As shown in Figure 1 The present application discloses a knowledge base knowledge association fusion method based on a knowledge graph; it includes the following steps: Step 1: Integrate structured, semi-structured, and unstructured data through cross-modal alignment technology to construct a multi-source heterogeneous data association network in a unified semantic space; Specifically, through cross-modal alignment technology (such as the CLIP model), visual, language, and voice heterogeneous data are mapped to a unified semantic space, eliminating image-text bias and audio-video description ambiguity, constructing a multi-source heterogeneous data association network to support intelligent question-answering accuracy improvement. Through knowledge abstraction modeling (RDF triples, attribute graphs) and embedding representation (low-dimensional vectors), cross-disciplinary and cross-language knowledge transfer is achieved, shortening the development cycle of multi-field applications.

[0022] Step 2: Use a dynamic attention mechanism to real-time fuse time-series data (acquire new external data) into the multi-source heterogeneous data association network, combine rule-based reasoning with machine learning models to resolve entity conflicts and solve cross-system semantic conflicts.

[0023] Specifically, a dynamic attention mechanism (such as the MCAN model) is used to real-time fuse new knowledge into the knowledge graph, combining rule-based reasoning with machine learning to detect and resolve entity conflicts; Step 3: Based on graph neural network, the implicit association between entities in the multi-source heterogeneous data association network is mined, and the logical chain of the knowledge graph is completed; Specifically, based on graph neural network (GAT), multi-hop neighbor relationship is mined, implicit association is discovered, missing logical chain in knowledge graph is completed, recommendation system accuracy is optimized, and upper layer knowledge fusion application precision is improved.

[0024] Step 4: Based on knowledge graph, hybrid retrieval architecture (BM25+vector+RAG) and reasoning engine support intelligent question answering and risk assessment decision-making scenarios; Step 5: Incremental learning technology is used to automatically update and expand the knowledge base of the association network knowledge graph, so as to deposit enterprise experience and drive knowledge base update.

[0025] Further, step 1 specifically includes the following steps: Step 1-1: Acquire semi-structured and unstructured data to extract knowledge and obtain knowledge extraction results; wherein, knowledge extraction includes entity extraction, relationship extraction between entities, and attribute extraction of entities; Step 1-2: Integrate the knowledge extraction results with structured data to obtain knowledge representation and eliminate information silos.

[0026] Step 1-3: Align entities and correct attributes of the knowledge representation to construct a multi-source heterogeneous data association network in a unified semantic space.

[0027] Further, the cross-modal alignment technology in step 1 uses the CLIP model to map visual, language, and speech data to a unified semantic space, eliminating discrepancies between text and image descriptions and inconsistencies between audio and video.

[0028] Further, the heterogeneous data in step 1 includes visual data, language data, and speech data, and the association network of multi-source heterogeneous data is constructed by eliminating discrepancies between images and text and ambiguities in audio and video descriptions.

[0029] Further, in step 2, the dynamic attention mechanism uses the MCAN model to detect and integrate new knowledge in real time, automatically trigger conflict resolution processes, solve synonymous entity naming conflicts or ambiguous relationship ambiguities, and ensure the timeliness of the knowledge base.

[0030] Further, in step 3, based on graph attention network (GAT), multi-hop neighbor relationship is analyzed, and knowledge completion suggestions are generated to optimize the accuracy of the recommendation system.

[0031] Further, in step 4, the hybrid retrieval architecture specifically includes: combining traditional retrieval (BM25), vector similarity calculation, and retrieval enhancement generation (RAG) model to improve the recall rate and accuracy of complex queries.

[0032] Further, asFigure 2 As shown, the steps of the intelligent question answering in step 4 are as follows: Step 4-1, obtaining the user's question and extracting keywords after query analysis, and vectorizing the keywords of the user's question; Step 4-2, retrieving a plurality of recalled knowledge items from the vector database of the knowledge graph by the vectorized keywords; Step 4-3, reordering the plurality of recalled knowledge items according to the correlation degree from high to low; Step 4-4, generating a question answer according to the recalled knowledge items and returning it to the user. Specifically, the question answer is generated by a large language model to present a natural language answer to the user.

[0033] Further, the incremental learning in step 5 includes: developing an association network expansion algorithm to automatically discover new entities and new relationships, and embedding and fusing the new knowledge into the knowledge graph through knowledge abstraction modeling (RDF triple, attribute graph) and embedding representation (low-dimensional vector).

[0034] Specifically, as a feasible implementation, the new knowledge is uploaded to the knowledge graph server in the form of a document, the server generates a corresponding parsing task for the uploaded document, and the uploaded document is parsed / OCR / layout analysis / table recognition through a related executor to obtain a vectorized knowledge representation, and the vectorized knowledge representation is embedded and fused into the vector database of the knowledge graph The present application generates a unified, structured and semantic knowledge network by associating and fusing heterogeneous data scattered in different knowledge bases through knowledge graph technology. The system includes a data management module and an algorithm capability module, and the algorithm capability module is further divided into a plurality of sub-modules such as model training methods. By introducing vector representation, clustering methods, graph neural networks and other advanced technologies, the accuracy and efficiency of knowledge fusion are significantly improved.

[0035] The present application adopts the above technical solutions, integrates multi-source heterogeneous data (such as documents, speech, images) and constructs a dynamic knowledge graph, uses cross-modal alignment technology and hybrid retrieval architecture to generate a knowledge graph, embeds the knowledge graph into an enterprise-level knowledge base reasoning process, constructs a dynamic semantic association network, supports knowledge fusion applications, and solves the problems of information silos and knowledge fragmentation. At the same time, the present application integrates multi-modal data across systems and in heterogeneous formats, mines implicit associations based on graph neural networks, supports entity alignment, attribute correction, semantic disambiguation and other functions, realizes continuous updating of knowledge assets and an intelligent decision-making closed loop, solves the problems of information silos, knowledge fragmentation and update lag, and supports intelligent decision-making and continuous value-added of knowledge assets in the scene.

[0036] The knowledge graph supports natural language question answering, and accurate answers are generated through knowledge reasoning, which significantly enhances user experience. In the recommendation system, the technology utilizes the entity correlation and reasoning capability of the knowledge graph to realize personalized recommendation.

[0037] Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. The embodiments in the present application and the features in the embodiments can be combined with each other without conflict. The components of the embodiments of the present application described and shown in the drawings herein can be arranged and designed in various different configurations. Therefore, the detailed description of the embodiments of the present application is not intended to limit the scope of the claimed application, but only represents selected embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the present application.

Claims

1. A knowledge graph-based knowledge base knowledge association and fusion method, characterized by: It includes the following steps: Step 1: Integrate structured, semi-structured and unstructured data through cross-modal alignment technology to construct a multi-source heterogeneous data association network with a unified semantic space; Step 2: Use dynamic attention mechanism to integrate new external data into the multi-source heterogeneous data association network in real time, and combine rule reasoning and machine learning model to resolve entity conflicts. Step 3: Based on graph neural networks, mine the implicit relationships between entities in the multi-source heterogeneous data association network to complete the logical chain of the knowledge graph; Step 4: Based on the knowledge graph, a hybrid retrieval architecture and inference engine are used to support intelligent question answering and risk assessment decision-making scenarios; Step 5: Use incremental learning techniques to automatically update and associate the knowledge base of the knowledge graph with the knowledge graph.

2. The knowledge graph-based knowledge base association and fusion method according to claim 1, characterized in that: Step 1 specifically includes the following steps: Step 1-1: Obtain semi-structured and unstructured data for knowledge extraction to obtain knowledge extraction results; among which, knowledge extraction includes entity extraction, relationship extraction between entities, and attribute extraction of entities; Steps 1-2 involve integrating the knowledge extraction results with structured data to obtain a knowledge representation, thereby eliminating information silos. Steps 1-3 involve entity alignment and attribute correction of the knowledge representation to construct a multi-source heterogeneous data association network with a unified semantic space.

3. The knowledge graph-based knowledge base association and fusion method according to claim 1, characterized in that: In step 1, the cross-modal alignment technology uses the CLIP model to map visual, linguistic, and speech data to a unified semantic space, eliminating text-image description bias and audio-visual inconsistency.

4. The knowledge graph-based knowledge base association and fusion method according to claim 1, characterized in that: In step 1, the heterogeneous data includes visual data, language data, and speech data. Before constructing the association network of multi-source heterogeneous data, the image-text bias and audio-visual ambiguity are eliminated.

5. The knowledge graph-based knowledge base association and fusion method according to claim 1, characterized in that: In step 2, a dynamic attention mechanism is used to detect and integrate new knowledge in real time using the MCAN model, automatically triggering a conflict resolution process to resolve naming conflicts of synonymous entities or ambiguities in polysemous relationships, ensuring the timeliness of the knowledge base.

6. The knowledge graph-based knowledge base association and fusion method according to claim 1, characterized in that: In step 3, multi-hop neighbor relationships are analyzed based on graph attention networks to generate knowledge completion suggestions to optimize the accuracy of the recommendation system.

7. The knowledge graph-based knowledge base association and fusion method according to claim 1, characterized in that: The hybrid retrieval architecture in step 4 specifically includes combining traditional retrieval, vector similarity calculation, and retrieval enhancement generative models to improve the recall and accuracy of complex queries.

8. The knowledge graph-based knowledge base association and fusion method according to claim 1, characterized in that: The steps for intelligent question answering in step 4 are as follows: Step 4-1: The server obtains the user's question, performs query analysis, extracts keywords, and vectorizes the keywords in the user's question. Step 4-2: Retrieve several recalled knowledge entries by searching the vectorized keywords in the vector database of the knowledge graph; Step 4-3: Reorder the recalled knowledge items from highest to lowest relevance. Step 4-4: Generate answers to questions based on the recalled knowledge items and return them to the user.

9. The knowledge graph-based knowledge base association and fusion method according to claim 1, characterized in that: Incremental learning in step 5 includes: developing an extended algorithm for the association network to automatically discover new entities and relationships, and embedding the new knowledge into the knowledge graph through knowledge abstraction modeling and low-dimensional vectors.

Images