Research On Point-of-interests Recommendation Algorithms In Location- Based Social Network

With the maturity of mobile positioning technology and the popularization of mobile terminals such as smart phones and smart bands,Location-based Social Networks(LBSN)has gradually become an indispensable part of people's life.In LBSN,the location recommendation service is also called Point-of-Interest(POI)recommendation.By analyzing the user's historical check-in data,POI recommendation mines user preferences,and predicts the locations that users may visit.This paper studies the application of traditional recommendation and in-depth recommendation methods to the POI recommendation.Based on the analysis of user check-in,a lot of research and improvements on the existing recommendation model are made.Two feasible improvement models are proposed,and the effectiveness of the model is verified through experiments.Firstly,a POI recommendation model that integrates the influence of locations is proposed.In order to alleviate the problem of data sparseness,2-degree friends are introduced into the collaborative filtering algorithm to obtain the influence of social factors on check-in.For low accuracy of the existing collaborative algorithm recommendation,a location influence model is built,combined with the kernel density estimation method in-depth,mining the influence of geographical location factors.At the same time,2-degree friends' check-in records are used to construct candidate sets,to improve recommendation efficiency.Secondly,a deep POI recommendation model based on multi-feature representation and Attention mechanism is proposed.On the basis of research on depth recommendation methods,we construct extraction model of the POI and user features.Specifically,the constrained matrix decomposition method is used to obtain the joint representation of the POI location and category features,and the word embedding model is used to obtain the semantic features of the POI,so to implement a comprehensive and in-depth multi-feature representation of the POI.Meanwhile,the user feature generation method is improved,and the Attention mechanism is used to improve the abilities of models to extract users' preference.In addition,the paper builds a deep learning recommendation framework used to model the nonlinear interaction between features to improve the accuracy of POI recommendation.Finally,on the basis of the above research,we design and develop an LBSN recommendation prototype system,completing the design of the system interface and business process.Then we encapsulate and reconstruct the models and methods proposed in the algorithm research process,to form an LBSN recommendation system for data analysis and display.In this thesis,there are 36 figures,5 tables and 84 references.