| With the global circulation of Internet information and the popularization of digital,networked,and intelligent devices,the development of social networks is also changing with each passing day.Among them,the point-of-interest(POI)recommendation problem in location-based social networks(LBSN)has attracted more and more public attention.LBSN can provide users with real-time location services,which is convenient for users to record the locations they are interested in and share their pleasant experiences with friends.POI recommendation recommends locations by mining and analyzing users' historical check-in data.It plays an important role in academic and commercial fields,such as big data mining,artificial intelligence,commercial promotion,and potential user screening.It also faces challenges in terms of data sparsity,implicit user feedback mechanism,and personalization of user preferences.In order to mine more valuable information in LBSN and make recommendations satisfy the needs of users,we make use of the location,category,time,space,and other information in the check-in data to propose a more reasonable recommendation strategy.User recommendation and group recommendation are proposed to improve the accuracy of recommendation and user satisfaction.The main work of this paper consists of two parts.First,in terms of personal recommendation.(1)This paper studies the influence of time aware on location recommendation,that is,users have different preferences at different time slots of the day,and we recommend POI for target users at any designated time slot of the day.(2)We found that the user's check-in records are not necessarily all positive.There are some "noisy locations" where the user has checked in but the evaluation is very low.In response to this phenomenon,this paper uses a tensor decomposition algorithm based on user-category-time to mine user category preferences,thereby filtering locations that do not belong to user preference categories,and reducing the impact of "noise location" on the recommendation results.(3)This paper uses the spatially constrained Hypertext Induced Topic Search(HITS)algorithm to constrain the candidate locations in geographic location,and constrains the number of similar users in user-based collaborative filtering,and proposes an integrated POI recommendation model of time,space,category and user relationship.Finally,the two real data sets NYC and YK of Foursquare are used to compare and evaluate the model proposed in this paper.Compared with other cutting-edge algorithms,the method proposed in this paper has a significant improvement in evaluation indicators such as accuracy rate,recall rate,and F-measure value.Second,in terms of group recommendation.(1)This paper studies the influence of the number of group members on the recommendation results of POI.It divides the groups into two-person groups and multi-person groups according to their size,and proposes different recommendation methods for the two different groups.(2)Different from using recommendation strategies to integrate individual recommendations,this article does not directly calculate the preference probability of the group members for each point of interest.It regards the group rather than a loose collection of individuals.In this paper,a group feature preference model is established by considering the influence of category and time factors,and then the divergence degree within the group is calculated according to the user's friend relationship,considering the integrity and difference of the group.(3)This paper designs a method of mining group access data from personal check-in data and dividing training set and test set.It simulates group check-in on two real personal data sets in foursquare,which solves the problem of scarcity of group check-in data sets.At the same time,we also compare with the other three group recommendation algorithms in different directions,and all the experiments are significantly better than other comparison algorithms. |
PDF Full Text Request