A knowledge representation learning method of R-GCN and transformer fusion network in information retrieval field

By integrating R-GCN and Transformer networks, the problem of insufficient global structural information capture in knowledge graph representation learning is solved, achieving more accurate knowledge representation and semantic understanding of information retrieval systems, and improving the performance of entity classification and path query.

CN119202266BActive Publication Date: 2026-06-26HARBIN INST OF TECH AT WEIHAI +1
View PDF 4 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
HARBIN INST OF TECH AT WEIHAI
Filing Date
2024-09-05
Publication Date
2026-06-26

AI Technical Summary

Technical Problem

In knowledge graph representation learning, existing technologies suffer from limitations in capturing global structural information through graph convolutional networks and in directly processing node position or structural information through Transformer models, leading to performance constraints on certain tasks.

Method used

We employ a fusion network of R-GCN and Transformer to construct an information retrieval knowledge graph. We utilize R-GCN to capture local dependencies in the graph structure and combine it with Transformer to dynamically capture contextual information of sequence data. By fusing the representation vectors of both, we achieve a more comprehensive knowledge representation.

Benefits of technology

It improves the accuracy and effectiveness of knowledge representation learning, enhances the semantic understanding ability of information retrieval systems, and performs particularly well in entity classification and path query tasks.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119202266B_ABST
    Figure CN119202266B_ABST
Patent Text Reader

Abstract

The application provides a knowledge representation learning method of an R-GCN and a Transformer fusion network in the field of information retrieval, relates to the technical field of knowledge graph representation learning, and comprises the following steps: constructing input features of entities and relations based on an information retrieval knowledge graph; constructing a Transformer branch representation vector of the entities, an R-GCN branch representation vector of the entities and a Transformer representation vector of the relations based on a Transformer model and an R-GCN model, and obtaining a final embedding representation of the relations; obtaining a fusion feature of the entities in the information retrieval knowledge graph as the final embedding representation of the entities based on the R-GCN branch representation vector of the entities and the Transformer branch representation vector of the entities; and optimizing the effect of knowledge representation learning through information retrieval knowledge graph entity classification and information retrieval knowledge graph path query answering tasks. The application captures local dependency relationships in a graph structure through a GCN, dynamically captures dependency information at different positions in sequence data through a Transformer model, captures global structure information, fully utilizes global structure information and context information in the information retrieval knowledge graph, and realizes more accurate knowledge representation learning.
Need to check novelty before this filing date? Find Prior Art