Knowledge Graphs in Recommender Systems

Abstract

Recommender systems are used to overcome the information overload problem and provide a personalized recommendation to the user. Recommender systems suffer from several challenges, like data sparsity and cold start problems. Knowledge graphs are proven to benefit recommender systems in multiple ways, like elevating cold start problems, tackling data sparsity problems, increasing the accuracy of recommendations, and can also be used to provide explanations for recommendations. Knowledge graphs are heterogeneous information graph that is made up of entities as nodes and relationships as edges; they store rich semantic information. Many researchers have used a knowledge graph for recommendations of movies, news, music, fashion, etc. Knowledge graph embedding or knowledge graph representational learning is the trending approach for the usage of knowledge graphs in recommender systems. Translational models like TransE, TransH are widely used and neural network models like graph convolutional network and graph attention network are seen to perform efficiently. The article, gives a background of recommender systems and knowledge graphs and then talks about some different approaches available for their integration and also presents a unique classification of available methods.