Research On POI Recommendation Model Based On Multiple Factors

The rapid development of mobile technology and location-based social network makes it easy for people to obtain various types of information resources.Users are willing to share personal experience in LBSN and hope to get accurate personalized services,so POI recommendation comes into being.This paper mainly studies POI recommendation and successive POI recommendation based on multiple factors.The main work and contributions are summarized as follows.Firstly,a POI recommendation model based on time-aware and social relationships is proposed,which is called TPR-TF.The traditional POI recommendation usually divides the time periods artificially evenly,and only considers the influence of neighboring friends in social relationships,which cannot meet the personalized needs of users.In this paper,a time dynamic partition algorithm based on hierarchical clustering is designed to meet the user's time-aware requirements,so that the results of POI recommendation can change dynamically over time.At the same time,the common influence of direct and potential friendships is added into TPR-TF model to further expand the scope of social relationships.In addition,in order to improve the performance of the model,a random selection method based on POI access frequency distribution is designed to improve the classical BPR method.The experimental results show that TPR-TF model has higher precison and recall than state-of-the-art POI recommendation models on real Gowalla and Brightkite datasets.Secondly,a POI recommendation model based on text semantics and image features is proposed,which is called TSIFP.Existing studies on POI recommendation often neglects multimedia information that can bring rich implict features,such as text and image.TSIFP model uses deep learning technology to obtain text semantics and image features associated with POI from text and image information,and uses probabilistic matrix factorization to combine their effects.It not only enhances users' preference understanding for POI selection,but also further alleviates the sparsity of data.At the same time,TSIFP model is a basic POI recommendation model based on multimedia information in social networks.It can combine other factors to obtain a variety of integrated POI recommendation model,which has strong scalability.The experimental results show that TSIFP model achieves better model performance than state-of-the-art POI recommendation models in real Instagram datasets.Thirdly,a successive POI recommendation model based on spatial,temporal and social relationships is proposed,which is called SpTeSo_SPOI.Different from POI recommendation,successive POI recommendation is more sensitive in time and space.SpTeSo_SPOI model uses spatial influence,temporal influence and social relationships to construct the model,so that the recommendation results can change with different time,POI location and social relationships.In order to enhance the learning ability of the model,two methods based on locational partial ordering and temporal partial ordering are designed to carry out personalized learning.The two methods can be applied to other successive POI recommendation models,which are universal.The experimental results show that the F1-score of SpTeSo_SPOI model on real Foursquare and Gowalla datasets is better than that of mainstream models,and the models obtained by using two partial ordering methods can achieve better recommendation results.Fourthly,a successive POI recommendation model USTI_SPOI is proposed,which combines user behavior sequence preferences with text semantics and image features.Most of the existing successive POI recommendation is based on sequence information,but there is no successive POI recommendation model that uses multimedia information such as text and image combined with sequence information.USTI_SPOI model uses two kinds of deep neural networks to obtain the long-term and short-term preferences of user sequence behavior,and the implicit user preferences in multimedia information.Finally,USTI_SPOI model is constructed based on the combined results of user implicit vector and POI implicit vector.The experimental results show that USTI_SPOI model is superior to many mainstream successive POI recommendation models on real Instagram datasets,and F1-score is significantly improved compared with other mainstream models.