Abstract: Point-Of-Interest (POI) recommendation is an important personalized recommendation function in Location-Based Social Networks (LBSN). This paper proposes a prediction point based recommendation algorithm. The algorithm learns the user's travel behavior according to the user history POI data distribution in the LBSN, and uses the variable Markov algorithm to predict the semantic information of the user's future arrival POI according to the current location. The final recommendation takes into account the difference in the number of user sign-ups and recommends N for the user. Highly interesting POI. The experimental results show that the proposed algorithm is higher than the two comparison algorithms in accuracy and recall rate, which indicates that the algorithm improves the recommendation effect of interest points and can effectively recommend the next interest points for users.