Research On The Spatiotemporal Data Mining-Based POI Recommendation

Point of interest(POI)recommendation is a fundamental task in location-based social networks(LBSN).The increasing proliferation of LBSNs brings about considerable amounts of user-generated check-in data.Such data can significantly contribute to understanding user behaviors,based on which personalized recommendations can be efficiently derived.POI recommendation plays an important role in encouraging users to explore cities and promoting the offline economy and has recently drawn extensive attention from academia and industry.According to different application scenarios and recommendation targets,POI recommendation problems can be divided into two categories:general POI recommendation(often abbreviated as POI recommendations)and next-POI recommendation.The POI recommendation aims to recommend all POIs that the user may be interested in,which are suitable for long-term planning scenarios.The next-POI recommendation aims to recommend the next POI immediately after the current checkin record,which is suitable for short-term planning scenarios.Users’ activities in the physical world are affected by travel distance and time.Therefore,spatial and temporal effects are two crucial factors in the user’s decision-making for choosing a POI to visit.Due to the heterogeneity and complexity of spatiotemporal contextual information,it is still challenging to effectively utilize such context information in POI recommendations.In this thesis,we propose recommendation algorithms for POI recommendation and next-POI recommendation by studying how to use spatiotemporal contextual information reasonably and effectively.Although most existing methods regard spatial and temporal effects as two independent features,in POI recommendation,spatial and temporal effects are highly interdependent.Therefore,this thesis proposes a spatiotemporal heterogeneous information network(HIN)based POI recommendation algorithm.Specifically,the main work and contributions are as follows:(1)This thesis proposes a HIN-based spatiotemporal joint modeling method.This method uses HIN as a modeling tool,and jointly models temporal information and spatial information as an entity in the HIN;(2)This thesis extends the traditional user-item binary interaction recommendation model and proposes a four-way interaction model that explicitly models spatiotemporal information.Experimental results on real-world datasets demonstrate the effectiveness of the algorithm in POI recommendation.Most of the existing next-POI recommendation research works are based on the recurrent neural network(RNN)to model the user’s historical check-in sequence.Limited by the sparsity of user check-in data,the hidden state of RNN is full of noise in the process of transmission.In addition,for the user’s current spatiotemporal context,the importance of the hidden state in different spatiotemporal contexts in the RNN is different.Most of the existing research works ignore this point.Therefore,this thesis proposes a sequential rule mining and spatiotemporal pattern search-based next-POI recommendation algorithm.Specifically,the main work and contributions are as follows:(3)This thesis proposes a new set of spatiotemporal sequence rules to solves the problem that the original sequence rule can only express sequential co-occurrence,but cannot be applied to check-in sequence enhancement.After that,three spatiotemporal search functions are designed to model human behavior patterns:temporal periodicity,time decay,and distance decay.Based on the spatiotemporal search functions,this thesis proposes the next POI recommendation algorithm that can search the more important neuron hidden states in RNN based on the current spatiotemporal context of the user.Finally,experimental results on real dataset show that the algorithm proposed in this thesis achieves 0.1738 and 0.2192 on Rec@5 and Rec@10,respectively,an improvement of 8.6%and 7.3%compared to the optimal baseline.