Enhancing Recommendation Accuracy and Explainability by Integrating Knowledge Graphs and Social Networks with Graph Attention
Year: 2025
Type: preprint
Abstract: This paper proposes a knowledge-aware influence propagation recommendation model based on graph attention, integrating user–item interactions with semantic information from knowledge graphs (KGs) and neighbor relations from social networks. In semantic propagation, user preferences are decomposed into deep interests captured via multi-step knowledge graph attention and shallow interests from historical interactions. Influence weights between items are derived from co-interaction frequency and semantic similarity, while user weights incorporate social ties. Final user and item embeddings are obtained through multi-layer propagation. Experiments on seven real-world datasets show that the proposed model outperforms strong baselines in recommendation accuracy. Moreover, we introduce a unified metric-based framework for explainability in multi-source recommendation, including three complementary metrics and a composite Total Score for balanced assessment. Results demonstrate that the proposed model achieves superior and consistent performance across all explainability metrics, attaining the best Total Score in all scenarios. To the best of our knowledge, this is the first work showing that jointly leveraging user-item interactions, KGs, and social relations improves both recommendation accuracy and explainability.
Authors: Xiantao Meng, Yuping Yang, Guifei Jiang
Institutions: Nankai University, China Mobile (China)
Language: English
Subfield: Information Systems
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
OpenAlex ID: W4414798301
Cites:
Cited by:
Related to: