A new important-place identification method

For the problem that the location of LBSN users was sparse because of it was updated sporadically. An important place identification method was studied by improving the clustering algorithm. Firstly, a velocity pruning algorithm was applied to remove interference points from source data. Secondly, clustering algorithm based on density peak named CFSFDP was selected. To solve the problem that it is unable to decide the cluster number with CFSFDP, an improved algorithm was presented. With a cluster center automatic choosing strategy, the algorithm searched for the `turning points´ with the trends of cluster center weights changing. Then we could regard a set of points which weight is bigger than turning points´ as cluster center. As a result, the error brought by ruling in the Decision Graph could be avoided with the strategy. Thirdly, reverse geocoding technology was applied to translate coordinate in semantic place. Check-in data of Foursquare user was set as an example in experiment, the result show that the improved algorithm had higher accuracy rate and lower computation compared to CFSFDP and DBSCAN. It also proved that the feasibility and effectiveness of the method proposed in the paper.