Feedback Knowledge Graph for Recommendation

Abstract

Knowledge graph is wildly used in recommendation system to deal with the sparsity and lack of interactive information. The negative samples in most recommendation systems are obtained by negative sampling from the no interactive data. There are some false marks in negative sampling, and lack reasonable explanation. The real negative examples should exist in the user's interaction history, and user's preferences can be obtained from the feedback data (i.e.: ratings, reviews). In this paper, we use feedback data to find the negative samples, and propose a Feedback Knowledge Graph (FKG) which can give good explanation for the results. Specifically, the interactive data is divided into positive and negative samples by feedback information, and attention mechanism is introduced to aggregate information from different neighbors. Finally, we use a neural network to make interactive prediction of user items. Extensive experiments on two real-world datasets show that our model achieves good results.