Multiple Similarities for Diversity in Recommender Systems

Compared to search engines, recommender systems provide another means to help users to access information. Recommender systems are designed to automatically provide useful items to users. A new challenge for recommender systems is to provide diversified recommendations. In this paper, we investigate an approach to obtain more diversified recommendations using an aggregation method based on various similarity measures. This work is evaluated using three experiments: the two first ones are lab experiments and show that aggregation of various similarity measures improves accuracy and diversity. The last experiment involved real users to evaluate the aggregation method we propose. We show that this method allows the balance between accuracy and diversity of recommendations. Keywords-Recommender System; Diversity; Similarity Measures; Users Study; Information Retrieval