Point-of-Interest Recommendation Based On Location-based Social Networks

In recent years,with the continuous development of Internet technology,Locationbased Social Networks(LBSN)has been widely used,and Point-of-Interest(POI)recommendation is one of its important services.Although POI recommendation has achieved certain research results in both academia and industry,there are still problems of sparse data and insufficient contextual information mining.The POI recommendation model is improved by deepening the influence of contextual information such as social,geographical and temporal information on users' check-in behavior based on existing research.1.A POI recommendation model that fuses social relationships and local geographic factors is proposed.To alleviate the impact of sparse check-in data,this thesis enriches the valid data by fusing social relationship and geolocation information.First,when analysing the influence of social relationships,considering that most users in LBSN have few or no check-ins with their social friends,it is impossible to accurately calculate the check-in similarity between users and social friends,this thesis reduces the calculation error by introducing the distance similarity.Second,when analysing the influence of geographical factors,this thesis divides a local activity area for each user,and more fully dissects the geographical preference of users' check-in according to the check-in situation of POI in the area.Finally,a POI recommendation model is constructed based on matrix decomposition fusing social relationships and local geographical factors.Experimental results show that the model improves in both accuracy and recall,and can effectively alleviate the data sparsity problem.2.A POI recommendation model based on the influence of multicentric spatiotemporal factors is proposed.At present,the research on the influence of temporal and geographic location context information in POI recommendation is often modeled separately for temporal and geographic location information,without digging deeper into the association between the information,which cannot model the user check-in behavior comprehensively and accurately and accurately.Therefore,we dig deeper into the spatiotemporal associations of user check-in by combining time and geographic location information.First,analyse the multicentric check-in behavior of users in different time states,and construct a polycentric geographic factor influence model in different time states.Second,this thesis also analyses the non-uniformity and continuity characteristics of users' check-in time,and models the influence of these two temporal characteristics on users' check-in behaviour respectively.Finally,a POI recommendation model is constructed based on matrix decomposition fused with spatio-temporal contextual information to better match user preferences.The experimental results show that the model can effectively tap the spatio-temporal contextual information to improve the recommendation performance.