Clustering Users' Pois Visit Trajectories For Next-Poi Recommendation

A novel recommender system that supports tourists in choosing interesting and novel points of interests (POIs) is here presented. It can deal with situations where users' data is scarce and there is no additional information about users apart from their past POIs visits. User behaviour is modelled by first clustering users with similar POIs visit trajectories and then learning a general user behaviour model, which is common to all the users in the same cluster, via Inverse Reinforcement Learning (IRL). Finally, recommendations are generated by exploiting the learnt behavioural models. The analysis of the produced clusters of trajectories and the generated recommendation shows that the proposed approach outperforms a baseline kNN model along several dimensions except precision.