Research And Application Of Point Of Interest Recommendation In Location Based Social Network

In location-based social network,people can share their position and activities to other users through check-in.Large volume check-in data provides an opportunity to mining user preference.Point of interest recommendation(POI recommendation)not only can help users identify favorite locations,but also can help POI owner acquire more target customers.POI recommendation has been a hot topic in academia and industry.As a branch of recommendation system,POI recommendation has been studied by lots of researchers,but existing research methods are basically from traditional recommendation.These methods didn't consider the complexity of the mobile environment and have some limitations.When analyzing user behavior in mobile environment,most studies only consider single element and ignore the collaborative effect of multiple factors.In order to solve these problems,we propose two personalized POI recommendation methods fusing multiple factors,the main work is listed as follows:(1)We analyze the research background,state-of-the-art,related theories of POI recommendation and their limitations.(2)Based on user-based collaborative filtering,we use TF-IDF method to represent user preference on locations and incorporate it into similarity.Then we analyze the effect of geographical distance and location popularity seldom considered.Moreover,we analyze the influence of social network to user behavior from social reputation,a new perspective of social network.All mentioned factors are integrated into a unified model named TSG.Experimental results on Brightkite dataset demonstrate TSG can improve POI recommendation effectively.(3)We utilize the seldom studied category information.The category of location is extraction of its function and semantics.User preference on location category can point user interest clearly than a specific location.Through analyzing Yelp dataset,we find the ratio of common visited category is higher than the ratio of visited locations.It means user preference on location category is more similar than user preference on location.Social regularization are used to model social relation.Finally,we propose a method based on location category and social network POI recommendation method named CSRS.Experimental results on Brightkite dataset demonstrate TSG can improve POI recommendation effectively.(4)Based on the proposed algorithms,we design and implement a personalized restaurant recommendation prototype system named CoolFeed.The system provides check-in,rating and reviewing restaurant,personalized recommendation and other functions.It mines user location preference and category preference from user check-in data,restaurant's position,restaurant's category and social network data,then generates personalized restaurant recommendation.