NILR:N-Most Interesting Location-based Recommender System

Location-Based Recommender Systems (LBRSs) have gained popularity in recent years as users tend to make decisions based on what are shared in social medias. Such systems depend on each user's historical behavioral information (or user profile) to determine users’ interests. However, it is impossible for new users to have the profile, making it difficult and challenging to recommend interesting locations (also known as a cold start problem). In order to tackle this issue, we propose an enhanced method, called N-most interesting location-based recommender system (NILR), which effectively recommends the N-most preferred places for each user without leveraging her profile. We also introduce a novel metric (so called interestingness score) to measure locations’ attractiveness. The metric takes into account both check-in frequencies and number of return visits of previous users already in the system. The method ranks the top-N locations based on the combination of the traditional HITS-based model (Hypertext Induced Topic Search) [1] and the proposed NILR. The results of the experiments on Foursquare dataset reveal that our proposed location recommender system and raking method perform effectively and efficiently, and outperform the HITS model in terms of accuracies and rankings.