The popularity of smart phones and the maturity of positioning technology spawn location-based social network (LBSN). The vast amounts of information assigns the location-based social network recommendation system a pivotal role. Current local point of interest recommendation algorithms often encounter two problems: synonymous site recognition and sparse data recognition. This paper proposes a local point of interest recommendation which regards the recommendation problem as a probability problem and uses categorical information to solve these problems, considering user preferences, time, distance and the popularity of a location during recommendation. Using Gowalla data, the experiment compares the proposed method to three other typical methods. The results indicate that our method has better performance and can provide a better user experience.