An improvement on recommender systems by exploring more relationships

Recommender systems are systems that can filter a great number of pieces of data and suggest mostly similar interested items of the user's preference. A variety of approaches have been proposed to perform recommendation, including content-based, collaborative filtering and association-based, etc. A potential problem existing in a recommender system is cold start [1]; simply defined that a system cannot draw any inference for users. In this paper, we deal with one of cold start problems by proposing a hybrid approach which combines two distinct features to solve the problem. While a user is related to other users in product purchase behaviors or preference, an item is connected to different items by its inside information. Our recommender system utilizes both these relations instead of each individual one to ameliorate the quality of output suggestion. This procedure will be revealed and discussed through this paper.