Abstract: With the pervasiveness of GPS-enabled smart phones, people tend to share their locations online or check in at somewhere by commenting on the merchants, thus arousing the prevalence of LBSN (location based social network), which takes POIs (point-of-interests) as the center. A typical application in social networks is the recommendation system, and the most common problem in recommendation system is cold start, that is, how to recommend for the users who rarely comment on the item or share comments. In this paper, we propose a recommendation algorithm based on circle and social connections in social networks. The circle is made up by all users who visit a particular category of items and their social connections. It means he is interested in this category that a user accesses the category of items. Our algorithm considers different social connections and circles on tradition matrix factorization. The social connections we use include the relationship between friends(explicit relation) and relevant experts(implicit), which are used as the rule to optimize the matrix factorization model. Experiments are conducted on the datasets from the 5th Yelp Challenge Round and Foursquare. Experimental results demonstrate that our approach outperforms traditional matrix factorization based methods, especially in solving cold-start problem.