Recurrent Knowledge Attention Network For Movie Recommendation

A primary concern of the current recommendation system is how to provide personalized recommendation to users and improve the accuracy and user satisfaction. The Knowledge Graph(KG) provides a new way to improve the recommendation system. This paper proposes a movie recommendation model based on Recurrent Neural Network(RNN) and KG─RKAN, which uses the auxiliary information in the KG to look for the potential interests of users for personalized recommendations. In addition, in order to solve the problem of user’s individual interests, an attention module was designed in RKAN,using different weights to converge user’s interest; multiple sets of negative samples were used for comparison to balance model training; data collected from the real movie data set Movielens and IMDB was mapped into a new data set for testing. Experiments show that the model has significantly improved the recommendation accuracy, and can better explain the reasons behind recommendations.