Book Recommendation Based on Joint Multi-relational Model

Recommender system, which is powerful to deal with the issue of information overload, has been widely investigated by many researchers recently. However, one of the biggest challenges needs to face is the cold start problem. To address this problem, the data source from social network is incorporated into our recommender system in this paper. In a social network, users who tightly connected imply some group-specific interests. Consequently, we may exploit social network information to resolve the cold start problem and improve prediction performance. The main motivation of this paper is to exploit social relationships and other extra data sources to adjust the latent factors learning over the target matrix, namely book rating matrix and a group of auxiliary matrices, typically, the social relationship matrix. Our recommender system is based on coupled matrix factorization in major, and utilizes the random walk and genetic algorithm to learn some special parameters. The data for experiments is crawled from one of the Chinese biggest reading-sharing website, Douban. Finally, the results have proved that our book recommender system incorporating auxiliary data sources has much better performance than traditional methods.