Research On POI Recommendation In Location-based Social Network

The rapid development of the fourth generation of mobile networks,social media and the ubiquity of the advanced mobile devices in which GPS modules are embedded enable location-based services,especially location-based social networks such as we chat.Therefore,POI(Point of interest)recommendation have received extensive research attention and a large number of researchers have proposed innovative and valuable POI recommendation algorithms.However,existing POI recommendation algorithms based on LBSN are affected by data sparsity,cold start and other problems,resulting in poor results.In this paper,the characteristics of location social network and some advantages and disadvantages of the existing representative POI recommendation algorithm are taken into consideration.The user's geographical location and time information are deeply mined and combined with the user's interests and hobbies to improve the accuracy of personalized location recommendation.The main research results are as follows:(1)In order to understand the advantages and disadvantages of the existing representative POI recommendation algorithms and the possible improvement direction,this paper proposes an algorithm analysis and improvement framework based on evaluation process.By composing the evaluation process layer,analysis layer and conclusion layer of the framework,the performance of seven representative POI recommendation algorithms in four scenarios,namely,different data sets,different data densities,different check-in numbers of users and different activity ranges of users,is compared,and the reasons are analyzed.According to the reasons,three possible improvement directions of the algorithms are proposed.(2)Combining real data set Gowalla,the relationship between time and geographical impact is analyzed.We draw three conclusions: in location-based social networks,users' check-in behavior also has geographic clustering phenomenon in time dimension;geographic clustering has continuity in time dimension;and geographic clustering has heterogeneity in time dimension.(3)Because of the relationship between time and geographical factor,we choose the matrix decomposition based LRT algorithm to be improved.By fusing the method of solving data density in GeoMF algorithm,we propose a POI recommendation algorithm based on time and geographical factor GeoTMF.GeoTMF is compared with LRT and GeoMF using Gowalla and Foursquare data in four scenarios.It is found that the performance of the algorithm is the best,and it can solve the data sparsity and cold start problems in POI recommendation to some extent.This paper makes a systematic study and summary of location-based social networks.Experiments on real data sets Gowalla and Foursquare show that the proposed GeoTFM algorithm significantly improves the accuracy of POI recommendation.At the same time,it also shows that the algorithm analysis and improvement framework based on evaluation process has certain reference and operability for the proposed new POI recommendation algorithm.