Research On Spot Recommendation Methods In Location-based Social Network

With the development of location-based services and social networks, both of them have great impact on our life.So combine their applications have become a very popular internet trends. Location-based social network provide a new platform for online social networking in virtual world and offline location information in physical world. The two of them can be fully integrated and play their strengths in platform. After collect a large number of user information, friend information, location information in this new platform, these data can be used to make personalized recommendation. Such as the activities recommendation and route recommendation. Spot recommendation is an important application in personalized recommendation. However,spot recommendation algorithm is not much. The problems such as data sparsity, cold start, check-in location away from residence are still not have good solutions.In order to solve the problems, spot recommendation in location-based social network is researched and an improved recommendation method is proposed. The user‘s interest preferences, friend relations, position semantic and distance and other factors was taken into account in this recommendation method. In order to solve the cold start and check-in location far away from residence problems, the method consider friend relations factor into user collaborative filtering algorithm. The user and friend relation collaborative filtering is proposed. In order to solve the data sparsity problem, the method based on the collaborative filtering algorithm and the clustering algorithm. The K-medoids division method based on semantic location and distance is used to cluster. The advantages of the two algorithms are complemented so as to solve the problem. To verify the proposed methods, precision, recall, mean average precision is used as measure on Foursquare dataset. The results indicated that the proposed method could solve common problems effectively and improve the recommendation effect.The focus of this research is the calculation of similarity. For the similarity between users. Using interest location similarity and friends intimacy to measure. The method find other users same to target user, and recommend their favorite places to target user. This can address issues such as cold start. For the similarity between sites. Using the term frequency- inverse document frequency and cosine similarity to measure position semantic similarity. Using spherical surface distance to measure the similarity of distance. Thereby cluster to locations and solve the data sparsity problem.