Research On Missing POI Restoration Algorithm For LBSN Based On Representation Learning

With the rapid development of mobile Internet and GPS positioning technology,mobile smart devices and location services have been widely used.The Location Based Social Network(LBSN)has gradually penetrated into every aspect of our lives.For example,Foursquare,Gowalla,Yelp,Instagram,Weibo and other platforms can find a large number of users’ check-in information.Location features shared by users and Point of interest(POI)information checked in LBSN connect social space with real space,making it possible to mine individual travel preferences and group travel rules from trajectory information.However,the lack of trajectory information in LBSN affects the quality of data and the relevant research on Points of Interest.What’s more,it is not beneficial to the understanding or mining of users’ travel behavior.The identification and restoration task of missing trajectory can be applied to the recommendation of Points of Interest.Meanwhile,it’s useful to the extraction of criminal clues in public security work.Most of existing methods for identifying and restoring missing points only rely on the attributes of the data itself,but the individual travel preference of user is a factor that cannot be ignored in the task actually.The failure to capture travel preference characteristics and spatiotemporal correlation greatly affects the accuracy of the results.Therefore,this thesis focuses on the problem of missing identification and restoration of LBSN trajectory.In our study,to better mine user travel preference and spatio-temporal correlation,two missing trajectory identification and restoration models based on representation learning are proposed,which effectively improves the accuracy of restoration.The detailed research work could be summarized as follows:(1)The real LBSN dataset is taken as the research object,and the analysis of its temporal and spatial characteristics is deeply analyzed,including the analysis of daily check-in mode,the analysis of time interval and space distance between continuous check-in points from the perspective of trajectory.Meanwhile,the analysis of time and space mode from the perspective of POI have also been analyzed,as well as travel transfer mode of users.The above information lays a good foundation for the following trajectory restoration model.(2)A Bi-direction LSTM Attention Fusion Model(Bi-LAF)for trajectory missing restoration is constructed.Firstly,the sequential information forward and backward the missing POI in LBSN trajectory was extracted,and user travel preference was mined using LSTM and self-attention mechanism.Secondly,user travel preference was further mined based on the time pattern of missing POI and user feature learning.Then,the complex relationship between missing POI and candidate POIs is captured by integrating the spatial and temporal information.Finally,the identification and restoration of missing POI was completed by modeling user travel preference and spatio-temporal complex relationship.The model is evaluated on two large scale real data sets,and the results show that our model has significant improvement over existing methods.(3)A bidirectional LSTM trajectory restoration model(E-Bi LSTMA)based on word embedding representation learning is proposed.The word embedding technology is used to capture the potential representation of Points of Interest.Then extracting the backward and forward trajectory sequence and the information of category sequence of missing POI according to the former result by Word2 Vec.Bi LSTM and attention mechanism are used to learn the potential features of sequences.Then the time pattern and user characteristics with missing POI are combined to mine user travel preference.Finally,the problem of identification and restoration of missing POI was completed combining spatio-temporal factors.The experimental results show that the prediction accuracy of the model is effective.