A Research On Point Of Interest Recommendation In Loation-based Social Networks

With the popularity of smartphones and the rapid development of mobile Internet, social networks are increasingly mature. Social network narrows people’s distance and reduces the cost of interpersonal communication. The maturity of positioning technology spawns location-based social network (LBSN). Compared to the traditional social network, location-based social network introduced geographic factors of a spot. Users can share their history records to others.But the vast amounts of information generated every day make it difficult for users to find the information they are interested in. In this trend, the location-based social network recommendation system is playing a pivotal role in improving users’ experience.There are three types of recommendation in LBSN:(1) Points of Interest (POI) recommendation:recommend sites to a user which he may be interested in (2) User recommendation:recommend popular users, potential friend or group to a user (3) Activities recommendation:according to the user’s interests, it is recommended that the activities he may participate in. POI recommendation is to be converted to an important part of 020 (online to offline), having a greater impact on user’s daily life, so this paper only studies point of interest recommendation.The current point of interest recommendation algorithm often encounter three problems:city recommendation,asynonymous sites recognition and data sparsity problem.This paper first describes the current mainstream algorithms of point of interest recommendation in location-based social networks and the challenges they faces.Then aimed at the three problems,it proposes concrete recommendation method in two different situations. The main contribution of the paper is as follows:(1) A local POI recommendation method is given which regards the recommendation problem as a probability problem that computes the user’s access probability to points of interest, considering local user preferences, time factor, distance factor and the popularity of a location.(2) A remote city POI recommendation method is given which regards remote city recommendation problem as a rating-score problem and recommend sites on the basis of user collaborative filtering recommendation. Firstly, the guide mechanism is used to select a series of guides in a city of each category, and calculate the similarity between users and different categories of guides.Then recommend spots according to the opinion of similar guides. In the model, the user’s preferences are represented by a curve named user category behavior timing curve, so the similarity calculation between the users is converted into the similarity between the curves.(3) During the recommendation, POI category information is fully used.In local POI recommendation method, the calculation of the probability of user u to visit locaiton v at a given time t is decomposed to the probability of user u to visit the category v belongs and the probability user u to visit location v among all the POIs belong to the categroy. In remote city POI recommendation, category information can help to predict the user’s preferences. Therefore, according to the category of local history records interest model of a user can be generated. So it can successfully resolving the problem of remote recommendation. Besides,the introduction of location category information transforms the user-location matrix to the user-category matrix, reducing matrix dimensions; on the other hand, based on the user-category matrix, the application of collaborative filtering can solve synonymous sites problem.In this experiment, the Recall@K is selected as an index to measure the recommendation quality. Using Gowalla data, the experiment compares the proposed method and the other three kinds of typical POI recommendation methods.The results show that whether it is in the local or remote scene, the point of interest recommendation method proposed in this paper has a better performance.