Combining Geographical And Social Influence In Deep Learning For Point-of-Interest Recommendation

Personalized point-of-interest(POI)recommendation plays an important role in location-based social networks(LBSNs).It not only helps users explore new places but also enables third-party services to better provide service such as targeted advertisements.Previous research studies on this topic mainly focus on check-in frequency but we argue that check-in frequency alone cannot entirely represent users' preferences.Moreover,most of the existing methods required careful engineering and considerable domain expertise to design a feature extractor,which is often very ineffective especially when the auxiliary information is very sparse.In this paper,firstly,we conduct a statistical analysis of check-in data and conduct a detailed analysis of geographical and social influence on users' check-in behavior.Moreover,we propose a POI recommendation algorithm based on deep-learning.More specifically,we use the semi-restricted Boltzmann machine to model the geographical proximity and the conditional layer to model the social influence.We propose a deep-learning method to transform both geographical and social influence into a high,abstract-level for POI recommendations.Our experimental results using datasets from real-world LBSNs show that our method achieves better performance than other state-of-the-art methods regarding three metrics,namely,precision,recall and F1.Conclusions of the paper are also given in the end: it is very important for LBSNs to improve POI recommendation service.Both geographical and social influences are essential to POI recommendation and these two influences can be modeled in deep-learning method to better achieve POI recommendation.