Striking the Balance Between Novelty and Accuracy in Location-Based Recommendation System

With widespread popularity of Location-Based Social Networks (LSBNs), the recommendation problem in this domain has led to significant amount of research regarding its practical applications. Despite extensive studies on recommendation systems based on parameters such as GPS trajectories, user-item ratings and check-in data, few methodologies take novelty of recommendations as a significant parameter. In this paper, we attempt to provide an improved approach to recommend points of interest (POIs) to users using a graph based approach which is in accordance with their personal interests and preferences. The proposed algorithm provides users with a personalized ranked list of venues based on their past check-in data and social relationships, which is novel yet accurate at the same time. It takes into account the existing challenges and is based on two key components: User Preferences and Social Relationships that are inferred from their past check-in history and Entropy of Venues which determines the novelty of recommendations provided. In short, it returns a ranked list of `k' venues which are most likely to suit the personal taste of the user. Experimental results demonstrate that the proposed methodology outperforms the baseline methods in terms of novelty and accuracy.