Prick the filter bubble: A novel cross domain recommendation model with adaptive diversity regularization

Recommender systems have been an important tool to filter and tailor the best content for online users. Classical recommender system methods typically face the filter bubble problem where users effectively get isolated from a diversity of viewpoints or content. How to provide relevant and diversified goods for online users has become a challenging problem. In this study, we develop a cross-domain matrix factorization model based on adaptive diversity regularization to address the above challenges. We leverage collective MF model to transfer users’ rating pattern, utilize social tags to transfer semantic information between domains, and design a novel adaptive diversity regularization to improve recommendation performance. Comprehensive experiments on real cross-domain datasets demonstrate the effectiveness of our model. Results show that our model can achieve a decent balance between recommendation accuracy and diversity, and the recommendation polarity can also be alleviated.