Meta-learning enhanced next POI recommendation by leveraging check-ins from auxiliary cities
Next point-of-interest (POI) recommendation has attracted a considerate amount of attention in the area of research recently to recommend the next POI where users are most likely to visit at the next time step. However, most existing next POI recommendation algorithms suffer from severe data sparsity issues, due to the scarcity of historical check-in data. Existing studies mainly resort to side information, such as POI categories, to mitigate the data sparsity problem, but ignores the rich check-in information from other cities. To this end, we explore how knowledge transfer from data-rich cities with diverse user patterns can help improve the next POI recommendation performance for cities with sparse check-ins. Accordingly, we propose a novel Meta-learning Enhanced next POI Recommendation (MERec) framework by leveraging check-in data from auxiliary cities, which incorporates the correlation of check-in behaviors among cities into the meta-learning paradigm. Concretely, the MERec framework takes into account the user check-in patterns of the target and auxiliary cities in terms of culture, urban structure, resident behaviour, etc., and transfers more relevant knowledge from more correlated cities. Extensive experiments on four real-world datasets demonstrate the superiority of our proposed MERec framework.