Multi-Effects Embedding Based Personalized POI Recommendation Method

With the rapid prevalence of smart mobile devices and the dramatic proliferation of location-based social network service(LBSNs),location-based recommendation has become an important means to help people discover new points of interest(POIs)and explore new regions.However,POI recommendation suffers from data sparsity problem,and even the travel locality makes it more challenging.There are many recent studies that exploit social effect,temporal effect,geographical effect,sequential effect,semantic effect etc.,to address data sparsity issue.However,they just exploit a part of above factors and lack a unified manner to integrated multi-effects.To address above challenges,we propose a generic graph and sequence joint embedding based POI recommendation system.our method jointly captures the social effect,temporal effect,geographical effect,semantic effect,user gender effect,user preferences,and sequential effect in a unified way by embedding the seven corresponding relational graphs(User-User,User-Time,POI-Time,POI-Region Hierarchy,POI-Category Hierarchy,User-Gender,and User-POI)and check-in sequences into a shared low dimensional space.To capture semantic information in check-in sequences,our method exploits sequence embedding model(word2vec),and others are incorporated by graph-based embedding method.Then jointly embedding learning above mult-effects using a joint training algorithm.Note that our proposed method is a generic flexible model that can be easily extended to incorporation of other factors.We conduct extensive experiments to evaluate the performance of our model on a real large-scale Foursquare dataset,and the experimental results show its superiority over other competitors.In addition,we study the contribution of each factor to improve location-based recommendation,and find that both temporal effect and semantic effect play more important roles than other effects.