Building Semantic Based Recommender System Using Knowledge Graph Embedding

Abstract

Recommendation systems are information filtering mechanisms used in E-commerce, media and entertainment industry. It essentially facilitate the customers for a better user experience by processing the content user-specific. This is known as personalization. However, though leveraged by machine learning algorithms existing recommendation systems, still suffers from the problem of cold-start and sparcity. These problems could be resolved by using knowledge graphs since it gives a semantic explanation of recommendations. Also, graph learning method overcomes the problems of manual feature extraction and is effective for feature learning in predicting tasks. In this research, we develop a semantic based recommender through link prediction in a knowledge graph. We apply graph embedding techniques for extracting the semantics of explicable recommendations. The proposed method is validated by building a knowledge graph using the MovieLens dataset. We observed that factorization based scoring functions such as HolE and DistMult provides better semantic recommendations.