A bi-section graph approach for hybrid recommender system

With the rapid growth of the World Wide Web, many efforts have been done to address the problem of information overload. Recommender systems help users make decisions in this huge information space. Most existing recommender system use either content-based or collaborative approach. It could be difficult to estimate a single best model for recommendation. Each of the single methods has own strengths, but also limitations and weaknesses. Therefore, combination of different methods can overcome these shortcomings and may result in better accuracy. In this paper, in order to have a better performance, we have introduced a hybrid recommender system which combines theses approaches in a bi-section graph model. We have gained web page similarity and user similarity by the new methods and have modeled web pages and users in the two-layer graph. Evaluation results show that combining these approaches achieved more accurate predictions and relevant recommendations than using only one of them.