Personalized Location Recommendation Algorithm Research Based On User Check-in Data

The popularity of smart devices,advances in broadband wireless networks and location sensing technologies,as well as the proposal of “Internet Plus” strategy lead to the emergency of location-based social networks(LBSNs).Huge amount of check-in data are generated in LBSNs.With the expansion of urbanization,the demand that recommends locations and routes for users becomes more and more urgent.In order to effectively help people find locations satisfying their preferences and improve the benifit of service providers,location recommendation based on users check-in data becomes an important task in LBSNs.In current location recommendation systems,the most common technique that was adopted by academia and industry is collaborative filtering.This technique does not consider the abundant geographical information and temporal information of check-in data,which results in the problem of data sparsity and poor recommendation performance.Moreover,for continuous location recommendations,or route recommendation based on user check-in data,most works do not consider both users' preferences and constraints on recommendations.Faced with these drawbacks,we fully take temporal and geographical information in check-in data into consideration,and propose location and route recommendation algorithms.The major works are listed in the following:1.In the paper,we propose a time-aware personalized location recommendation scheme.This scheme utilizes geographical information and temporal information,and considers user visiting pattern of periodicity,consecutiveness,and non-uniformness.At first,we extend the traditional user-location matrix to a user-location-time tensor to make use of temporal information.To relieve data sparsity problem,we construct user-user similarity matrix and location-feature matrix as supplementary information.Then,we apply tensor and matrix factorization to predict users' preferences for locations.At the same time,we adopt the idea of “split-and-conquer” and threshold-based algorithm to realize scalability.Finally,we conduct experiment on real-world Foursquare and Gowalla dataset to verify the effectiveness of our scheme.We test our recommendation algorithm with these datasets,and compare the experiment results with two existing location recommendation algorithms.2.For the cold-start problem that a user moves to a new city,we design a local feature aware location recommendation algorithm.In this recommendation algorithm,we take both users' preferences and local feature into consideration to solve the new city cold start problem.We first use user-based LDA model to learn users' preferences and use city-based LDA model to learn local feature of cities.Then,we combine users' preferences and local feature factors in an unified function,and the weight of users' preferences factor and local feature factor is computed by relative standard deviation function.The predicted users' preferences for locations are computed using this function.Finally,we conduct experiment on a real-world Foursquare dataset.3.We propose a time-aware personalized route recommendation algorithm.The route recommendation algorithm can better mine users' visiting pattern and users' latest check-in locations,and recommend routes under source location query and source-destination location query situations.We first construct user-location matrix based on check-in dataset and utilize non-negative matrix factorization to generate candidate location set.Then,we define a function that considers the popularity of lcoations,the proper visiting time of locations,the transition time duration between locations and proper visiting order of locations,to measure the quality of a route.At the same time,we propose route construction algorithm for source location query and source-destination location query,respectively.Finally,we conduct experiment on the public datasets of Gowalla and Foursquare.