Research On Location Recommendation Based Category Prediction In Location Based Social Networks

With the development of Internet and location technology,many location-based social networks(LBSNs)have emerged on the basis of traditional social networks.Large volume user data and location information in LBSN,which produce many location-based services,location recommendation is a typical representative.LBSN calls this location with service attribute as point of interest(POI).POI recommendation can help users find places of interest and improve their quality of life.The check-in behavior is affected by the user's current state and POI visited in recent period.This kind of POI recommendation problem,which focuses on the current state of users,has strong time constraints on recommendation results,the recommended POI are expected to be visited in the next period of time,is called Next POI recommendation.In the Next POI recommendation problem,category factors have a great influence on the recommendation results.The existing methods do not predict the category.When considering categories,the relationship between time and category is not well considered,lack of non personalized influence,many context factors are not well used.This paper divides the problem into two parts,predict the category,then recommend the POI,proposes a new Next POI recommendation algorithm based on category prediction.In this paper,based on check-in data,using tensor factorization and bayesian personalized ranking,we design a combined personalized and non personalized category prediction algorithm.On the basis of category prediction,combined with many context factors,two Next POI recommendation algorithms are designed.One mainly considers the influence of geographical location and popularity on Next POI recommendation.Another combines the predicted value of the category and designed Next POI recommendation algorithm based on the potential behavior and geographical influence.This paper improves the composition form of the existing recommendation list,designs a composition form combining the category prediction value.The experiments on real-world datasets demonstrate the effect of the algorithm.The experimental results show that the category prediction algorithm and the Next POI algorithm proposed in this paper have improved the recall rate and precisoin rate over several state-of-the-art methods.