Research On Point-of-Interest Recommendation Algorithms Based On Geographic Location And Social Connection

The widely popularity and application of location social services have led to Point-of-Interest(POI)recommendation systems to receive great attention from academia and industry.The POI recommendation is aimed to recommend locations which users are interested in,including shops,institutions,public places,etc.When users interact with the location service,a large amount of data is generated,such as reviews,ratings,social relationships,geographic location information,etc.These multi-source heterogeneous data provide a powerful support for mining user preferences.Through analyzing his-torical data,POI recommendation can not only benefit customers by helping them find places they prefer(exploring the city)and reduce their decision making time thus provid-ing satisfactory user experiences but also benefit merchants by increasing their revenue through virtual marketing.Apparently,exploring the rich multi-source heterogeneous information can effec-tively improve the accuracy of recommendation system.At the same time,it bring some challenges.On the one hand,user interest is diverse.On the other hand,similar to tra-ditional recommendation system,the POI recommender systems also face data sparsity and cold start problems,etc.Hence,How to use multi-source heterogeneous informa-tion to capture user interest diversity and to handle the data sparsity are the problems in this paper are designed to solve.This paper focuses on geographical location and social connection.In order to solve the above mentioned problems,this paper carries out research on POI recommendation technology based on geographic location and so-cial connection,that is,through the use of geographic location and social connection,users interests are accurately depicted,thus predicting and recommending the places that users will visit.Specifically speaking,the contents and achievements of this paper are as follows:Aiming at solving the diversity of user interests problem.this paper proposes a POI recommendation algorithm based on modeling user psychology.In general,user check-in behavior is a manifestation of user psychology.Specifically,users evaluations and visits to multiple places do not exist independently.Instead,these are the psychological performance of user preference after a comprehensive comparison of all locations in a mall.Based on the observation,the work is based on the check-in behavior of compara-bility,dissimilarity,and locality and exploits check-in data and location information to model user interest,thus can accurately describe the user psychological behaviors.This work combines utility theory in economics and power distribution to model user behav-ior.Specifically,we adopt utility theory and power distribution to depict comparability,dissimilarity and locality,respectively,and utilize utility theory to design optimization function,and then fuse power distribution to make final predictions.Extensive evalua-tion experiments conducted on Gowalla and Brightkite datasets confirm the superiority of our approach over state-of-the-art methods.Aiming at solving the data sparsity problem,this paper proposes a POI recommen-dation algorithm based on expert connection in a circle of interest.According to the "ho-mophily" theory in sociology,users have similar preferences to "friends" in their social networks,and several academicians exploit social connection as auxiliary information to improve recommendation accuracy.However,the existing recommender algorithms hardly explore the influence of opinion leaders underlying the social connections,which are described as experts.According to the theory of homophily,in this paper,opinion leaders in each circle tend to have greater impact on recommendation than those of friends with different tastes.Specifically,this work firstly uses unsupervised methods to identify explicit and implicit experts in social networks.And then users are divided into different interest circles according to the type of their visiting places,and users in each interest circle are influenced by experts in the field.Naturally,the regularizations of explicit and implicit experts are constructed to describe the influence of experts and learn user preferences.Experiments conducted on a real-world dataset demonstrate that our approach outperforms existing methods,particularly on handing cold-start problem.