Point of Interest (POI) recommendation systems exploit information in location-based social networks to predict locations that users may be interested in. POI recommendations have been widely adopted in many applications, which are helpful for daily life. POI recommendation services receive a huge volume of visit history data generated by users' daily lives with mobile devices. However, POI recommendation systems require long time to build a model from such a huge volume of check-in data and recommend suitable POIs to users. Thus, it is indispensable to shorten the execution time in a big data era. In this study, we propose a clustering-based method to divide the data into multiple subsets to accelerate the POI recommendation's execution while maintaining accuracy. Our proposed method can be adapted to any general POI recommendation algorithm. We divide the whole data, that is, users and POIs, into subsets with a tree structure to balance the size of subsets according to both geographical information and user check-in distribution. Evaluation results show that we successfully accelerate the base algorithms over 17 to 39 times faster while keeping the accuracy almost the same.