Subgraph Filtering Method of Knowledge Graph Based on Domain Attributes

As the size of knowledge graphs grows, directly using the complete graph consumes significant computing resources. To address this issue, various subgraph extraction methods have been studied. An improved subgraph filtering method based on domain properties was proposed, which designed to enhance the performance of recommendation systems through the extraction of specific domain subgraphs. This method focuses on the semantic distance between relation types and domain attributes, identifies domain relevance through iterative co-occurrence probability calculations of relations and related entities, and extracts relevant subgraphs using a weighted semi-supervised label propagation algorithm. Experiments on the Yelp 2018 dataset show that the extracted domain subgraphs maintain recommendation accuracy while reducing computation time by 75% compared to n-hop subgraphs, effectively enhancing recommendation system efficiency.