Building user Profiles Based on user Interests and Preferences for Recommender Systems

Nowadays, Web 2.0 allows users to express their opinions about items provided by various Internet services through reviews. These reviews towards particular items could contain a valuable information, such as the reasons why user like or dislike the item. This information can help more accurately describe a user for recommender system and, by that increase the accuracy of recommendations. Moreover, such reasons can reflect user preferences. However, extraction the user preferences from reviews is still a quite challenging task as reviews represented in the free text form. In this paper, a personalized recommendation approach based on user profiles of interests and preferences is proposed. In order to increase the accuracy of personal recommendations, user interests are obtained from item general description attributes such as item tags or categories, whereas user preferences are extracted from users' reviews. To evaluate proposed user profiles, we provided the top personalized recommendations for users. The experiments are conducted on a real-world dataset. We select candidates for recommendations based on user interests profile, and then, narrow and sort them by taking into account user preferences profile. We compare the results with several stateof-the-art recommendation algorithms using precision, recall and F1 metrics for evaluation. The results show the effectiveness and efficiency of our proposed approach in comparison with different recommendation algorithms.