A multi-factor influencing POI recommendation model based on matrix factorization

How to make recommendation for personalized users by using the available sparse data is a hot research topic in the area of big data and has wide application prospects. In this work, we investigate the POI (Point of Interest) recommendation of LBSN (Location Based Social Network) to provide users with personalized POI preference, such as attractions, hotels and shops and so on. A new POI recommendation model based on matrix factorization by considering the influences of both the geographical factor and the user factor, namely GeoUMF (Geographical and User Matrix Factorization), has been proposed in this paper. In GeoUMF, the objective function considers the difference between the ranking produced in the recommendation model and the actual ranking in the check-in data. In addition, an approximation method that considers the difference of visiting frequency of POI is defined in the objective function. Experimental results on real world LBSN data set demonstrate that GeoUMF obtained better performance in terms of the recommendation precision and the recall rate compared with some state-of-the-art algorithms in the literature.