Time Distribution Based Diversified Point of Interest Recommendation

In location-based social networks (LBSNs), personalized point-of-interest (POI) recommendation helps users mine their interests and find new locations conveniently and quickly. It is one of the most important services to improve users' quality of life and travel. Most POI recommendation systems devoted to improve accuracy, however in recent years, diversity of POI recommendations, such as categorical and geographical diversity, receives much attention because a single type of POIs easily causes loss of users' interest. Different from previous diversity related recommendations, in this paper, we focus on visiting time of POI-a unique attribute of the interaction between users and POIs. Users usually have different active visiting time patterns and different frequently visiting POIs depending on time. If a set of proper visiting times of recommended POIs concentrates on a small range of time, the user might be unsatisfied because they cannot cover whole of the user's active time range that results in inappropriateness for the user to visit those POIs. To solve this problem, we propose a new concept-time diversity and a time distribution based recommendation method to improve time diversity of recommended POIs. Our experimental result with Gowalla dataset shows our proposed method effectively improves time diversity 25.9% compared with USG with only 7.9% accuracy loss.