Personalized POI recommendation using deep reinforcement learning

Abstract

As an important location based service, next Point-Of-Interest (POI) recommendation has been widely utilized in helping people discover attractive and interesting locations. However, the sparsity of check-in data, the cold start issue and complicated contextual and semantic relationships between users and POIs bring severe challenges. To cope with these challenges, we develop a novel deep reinforcement learning based personalized POI recommendation framework. Within the proposed framework, a joint graph embedding model is proposed to compute users’ dynamic preferences, accounting for inter-user relationships, historical check-in sequence, and category information of visited POIs. We are working on the experiments of applying the proposed POI recommendation framework on two typical real-world location-based social network datasets.