Knowledge graph completion method based on granularity-aware graph attention network

By using a granularity-aware graph attention network approach, we can explicitly model global semantic information and dynamically allocate computational resources, which solves the problem of insufficient utilization of global semantic information in existing technologies. This improves the accuracy and interpretability of knowledge graph completion and optimizes computational efficiency.

CN122174956APending Publication Date: 2026-06-09HENAN NORMAL UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
HENAN NORMAL UNIV
Filing Date
2026-02-04
Publication Date
2026-06-09

AI Technical Summary

Technical Problem

Existing knowledge graph completion technologies suffer from insufficient utilization of global semantic information, lack of guidance from global semantic information in semantic aggregation, and rigid allocation of computing resources, resulting in insufficient reasoning accuracy, poor interpretability, and difficulty in balancing efficiency and performance.

Method used

We employ a granularity-aware graph attention network approach. By explicitly modeling global semantic information and dynamically configuring computational resources, we achieve information-guided semantic aggregation, generate a quantified entity-relation granularity weight matrix Q, dynamically select attention heads, enhance the embedding representation of entities and relations, and use a pre-defined scoring mechanism to predict triples.

Benefits of technology

It improves the reasoning accuracy and interpretability of knowledge graph completion, optimizes computational efficiency, achieves high performance on knowledge graphs of different densities, and demonstrates excellent multi-hop reasoning and complex logical inference capabilities.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122174956A_ABST
    Figure CN122174956A_ABST
Patent Text Reader

Abstract

The present application relates to the field of artificial intelligence and knowledge graph, and particularly relates to a knowledge graph completion method based on granularity perception graph attention network, based on the topological structure of the knowledge graph, mining the global semantic community formed by the shared relationship mode between entities, generating a quantitative entity-relation granularity weight matrix Q; according to the relationship distribution characteristics and semantic fineness of the knowledge graph, dynamically evaluating the importance of the attention head, screening out the target attention head subset to adapt to knowledge graphs of different densities; using Q to guide neighbor information aggregation, converting granularity semantic prior into edge weight, and fusing it with node features to generate enhanced entity and relation embedding representation; based on the enhanced entity and relation embedding representation, evaluating the effectiveness of the candidate triple through a preset scoring mechanism, and outputting the scoring result to complete the prediction and completion of the missing triple of the knowledge graph. The present application realizes the collaborative improvement of the accuracy, explainability and efficiency of knowledge graph completion reasoning.
Need to check novelty before this filing date? Find Prior Art