A Preference-based User Similarity to Construct a Collaborative Recommender System

Recommender systems have emerged as indispensable tools to overcome the information overload on the web. These systems provide personalization for users to select favorite resources. Collaborative Filtering is among the popular approaches is widely used to construct recommender systems. The main challenge of this approach is the effectiveness of its similarity measure to find similar users/items for providing recommendations. In this paper, we propose a novel measure to compute the similarity considering users'' preferences. The proposed similarity measure improves the performance of recommender systems by incorporating high level latent factors instead of co-rated items by users, or common selection patterns. These latent factors are extracted via preference-based user modeling. It also provides a better recommendation when only a few ratings (sparse dataset) are available for similarity calculation. We implemented the proposed method and evaluated it using the MovieLens dataset. The evaluation results show that the proposed preference-based similarity measure considerably improves recommendation performance compared to other existing approaches.