Multiple Factors-aware Point-of-interest Recommendation In Location-based Social Networks

Point of interest(POI)recommendation is a new form of popular recommendation in location-based social networks(LBSN).Utilizing the rich information contained in the LBSN to do personalized recommendation can enhance user experience effectively and enhance user's dependence on LBSN.Now researchers are facing the challenging problems in LBSN,such as no explicit user preferences,non-consistency of interest,the sparseness of data,and so on.POI recommendation is becoming research focus currently.Firstly,the traditional recommendation technology and the corresponding interest recommendation algorithm research are analyzed,and points out that the existing research has not considered fully.Then we point out the algorithm in detail according to the static recommendation of POI and real-time recommendation of POI respectively.Aiming at the static recommendation,a time-topic-aware POI recommendation strategy is proposed,of which,on the one hand,the historical check-in information of each user is divided into 24 time periods in hours;on the other hand,each POI is divided into a number of potential topics and distribution.Both the information of user's check-in and comments are used to mine user's topic preference in different time periods for Top-N recommendation of the POI.In order to achieve the recommendation ideas,first of all,according to the comments information on the visited POI,we use LDA topic generation model to extract the topic distribution of each POI.Secondly,for each user,we divide each user's check-in data into 24 time periods,and connect it with the topic distribution of the corresponding POI to map user interest preference on each topic in different periods.Finally,in order to solve the issue of data sparse,we use higher order singular value decomposition algorithm to decompose the third-order tensor of user-topic-time,to get more accurate interest score of users on each topic in all time periods.The experiments on a real dataset show that the proposed approach outperforms the state-of-the-art POI recommendation methods.Aiming at the real-time recommendation,a time-location-aware POI real-time recommendation strategy is proposed.It fuses multiple factors consisting of time-aware user characteristic,POI feature,users' personalized location transition preference and LBSN overall location transition and time characteristic,so it can be more accurate to obtain the user preference aware on the current time and location.First,we get the users characteristic matrix and POI feature matrix at different time periods by matrix decomposition to obtain the users' initial preference.Secondly,in order to get the time-aware relationship of every interest point at two levels,we achieve the personalized interest point transfer matrix of every user and the overall POI transfer matrix through the LT model.Finally,the factorization machine is introduced to build the TL-FM model.Not only it can integrate all the features,but also it generates the efficacious relationship between pairwise characteristics,to obtain the user preference which is aware on current time and location and achieve effective real-time recommendation.The experiments on a real dataset show that the proposed approach outperforms the state-of-the-art POI recommendation methods.