A Recommendation Model Fusing Knowledge Graph Convolutional Networks and Social Relationships

Abstract

Listen

Translate article

The knowledge graph’s rich semantic information can solve some of the issues with conventional recommendation. Most current knowledge graph-based recommendation models solely take into account the prospective characteristics of items, ignoring the supporting function of user potential features for recommendation systems, despite approaches like data sparsity and cold start. We suggest a recommendation model that incorporates social relationships and knowledge graph convolutional networks to solve this issue. Once more, the KGCN model is utilized to get the item feature matrix. Next, the multilayer perceptron is fed with the user feature vector, user embedding representation, and item feature vector to determine projected scores and provide user recommendations. The proposed recommendation model outperforms other benchmark models, according to experimental results that were done on the publicly accessible datasets CiaoDVD, Epinions, and FilmTrust to verify the model performance.