一种基于CFSFDP改进算法的重要地点识别方法研究  

Research on important places identification method based on improved CFSFDP algorithm

作　　者：马春来[1] 单洪[1] 马涛[1] 朱立新[1] 

机构地区：[1]电子工程学院,合肥230037

出　　处：《计算机应用研究》2017年第1期136-140,共5页Application Research of Computers

基　　金：国防重点实验室基金资助项目(9140C130104)

摘　　要：为解决CFSFDP聚类算法由于无法自动选择簇中心点而难以应用于重要地点识别的问题,引入一种簇中心点自动选择策略对算法进行改进。该策略将簇中心点权值的变化趋势作为自动划分簇中心的依据,有效避免了通过决策图判决簇中心点的方法所带来的误差。将CFSFDP改进算法与数据预处理及逆向地理编码等技术结合起来,能够以较高的精度实现重要地点识别。实验以Foursquare数据为例,结果表明CFSFDP改进算法比DBSCAN具有更高的准确率和较低的计算量,进一步证明了该方法在处理稀疏位置数据的重要地点识别问题上具有一定优越性。To solve the problem that CFSFDP clustering algorithm could not be applied to important places identification for the reason that it was unable to decide the cluster number with CFSFDP. This paper introduced a cluster center automatic choosing strategy to improve the algorithm. The strategy regarded the trends of cluster center weights changing as a rule with which decide the cluster center points automatically, avoiding the error brought by decision graph method. As a result, the method combing with CFSFDP algorithm, data preprocessing and reverse geocoding technology could improve the accuracy of important identification. The experiment chose Foursquare data as an example, the result shows that improved algorithm has higher accuracy rate and lower computation compared to DBSCAN. It also proves that the method has the advantage of other methods in handling the problem that important places identification with sparse location dataset.

关 键 词：重要地点识别 速度剪枝 基于密度的聚类 密度峰值 簇中心 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]