Research On Personalized Recommendation Based On User Influence And Content Aware Of Location-based Social Networks

With the development of the mobile web and smart devices, location-based social networks(LBSNs) came into being. It realized the effective interaction between the online virtual society and the offline physical world by introducing the location check-in mechanism into traditional social networks, thereby promoting the consumption of urban locations. However, the increasing number of users and locations has caused the problem of information overload. Therefore, studying efficient location recommendation based on the platform of LBSNs is an effective means for users to quickly find points of interest in massive amount of data.The sparseness of user check-in data has significantly limited the accuracy of current location recommendation technologies based on collaborative filtering; in the meantime, there is also a serious lack of research on location recommendation service for users who travel to remote cities. For this problem, based on the study of the influence of users, the content matching of user interests and location information, distance and attractiveness of location, this article proposed the UCFTCL(User-CF-Trust-Content-Location) location recommendation algorithm which achieved effective location recommendation in both local and remote city. This thesis has theoretical and practical significance, and the main contents are as follows:(1) It analyzed the advantages and disadvantages of current location recommendation algorithms of LBSNs.(2) It used collaborative filtering algorithm to analyze the behavior attendance similarity between users, and introduced social trust to propose UCFTR(User-CF-Trust) algorithm. It also analyzed the location recommended efficiency under the combined effects of user's check-in data and social relationships.(3) For the problem of the diversity of tags of the same location and theme similarity of different positions, this thesis introduced interest-feature contrast analysis to propose the improved user influence calculation algorithm UCFTRI(User-CF-Trust-Interest).(4) When users travel to remote cities, the social impact is weakened, and attendance data is sparse. For this problem, this thesis used LDA(Latent Dirichlet Allocation) topic model to extract user's interest features, taking off-site urban theme preferences into consideration, and analyzed the content matching degree between user interests and location information. It combined the factors of user influence, content matching, distance and attractiveness of the location to propose location recommendation algorithm UCFTCL, implementing the adaptive location recommendation according to user's current city. This thesis verified the validity of this algorithm through experiments on Foursquare data set.(5) On the basis of UCFTCL algorithm, this thesis used the open data interface of Sina microblogging to design and implement a personalized location recommendation system prototype based on LBSNs.