Location-Time-Aware Travel Attraction Recommendation In LBSN

With the development of the tourism industry, more and more tourists become online travel planners. People usually want to take both hot tourist attractions and personal preference into consideration while choosing attractions. Then, how to quickly and efficiently help users personalized selection of scenic spots is the right question which urgently to be solved.At the same time, location-based social network platforms (LBSNs) represented by "Foursquare" got rapid development. LBSNs manage to link offline physical world with the online virtual world through functions like "check-ins". Abundant social, time and space information is contained in LBSNs, which undoubtedly provides an excellent data source for mining user interset and making personalized recommendation.Although there have been some studies in personalized attraction recommendation, there still remains some unsolved problems in modeling user behaviors without taking time and spatial information into consideration. Besides, there are still some problems, such as "data sparsity problem" and "recommendation for out-of-town users".This paper carries out the research of personalized attraction recommendation based on checkin data and social data in LBSNs. The research considers the time and spatial features in user behaviors to solve problems mentioned before.Firstly, this paper proposed an attraction checkin dataset extraction method based on density clustering method. This method managed to extract attraction checkin dataset and preliminary alleviate the data sparsity in LBSNs at the same time. Secondly, this paper introduce a kind of latent factor model, which called LDA topic model, to describe user interest, according to the fact that collaborative filtering has poor performance dealing with sparse data. Then this paper proposed a spatial expand LDA model called LLDA by taking geographical influence into consideration. The training algorithm for LLDA model is also given at the same time. Based on the these work, we further design a Iocation-time-social aware recommendation method called LTSR which takes several factors such as user preference, social relationship, local hotspot and time influence into account. Finally, a personalizd attraction recommendation system is designed and implemented based on the LTSR recommendation scheme proposed in this thesis.Besides, comparison experiments based on the Foursquare dataset is conducted in order to present the effectiveness of LTSR method. According to the fact that attraction recommendation always involves new city situations, this paper designed two cases of the experiment, recommendation for in-town-users and recommendation for out-of-town user.The experimental results show that our approach significantly outperforms existing recommendation methods in the effectiveness of both in-town and out-of-town recommendation.