主动地纠错式半监督聚类社区发现算法-ChinaXiv.PDFVIP

/article/02-2019-09-029.html 计算机应用研究 主动地纠错式半监督聚类社区发现算法 * † 张贤坤，刘渊博，任 静，张高祯 (天津科技大学 计算机科学与信息工程学院, 天津 300457) 摘 要：经典的无监督聚类算法快速、简单且可以直接对大规模数据集进行划分，但是由于网络结构较为复杂，划分 ESCD error correction semi-supervised 的准确度并不高。为此，提出一种基于主动学习的纠错式半监督社区发现算法 （ community detection algorithm ），将传统的K-means 算法进行分步计算，并且在聚类的过程中加入成对约束。根据先验 信息保留正确的划分，纠正错误的划分来改变网络的连接关系，使网络具有更明显的块结构，当节点与聚类中心的距 离不再变化时划分结束。实验结果表明，与现有的社区发现算法相比，ESCD 算法具有更高的精度，且所需的监督信 息远远小于其他半监督算法。 关键词：主动学习；纠错式半监督社区发现；K-means 算法；成对约束 中图分类号：TP301.6 doi: 10.3969/j.issn.1001-3695.2018.03.0162 1 v Active error-correcting community discovery algorithm based on semi-supervised clustering 3 6 † 4 Zhang Xiankun, Liu Yuanbo , Ren Jing, Zhang Gaozhen 0 (School of Computer Science Information Engineering , Tianjin University of Science Technology, Tianjin 300457, China) 0 . 5 Abstract: The classical unsupervised clustering algorithm is fast, simple and suitable for mining large-scale datasets, and it can 0 also directly divide communities. However, due to the complexity of communities, the classification accuracy of the algorithm 8 1 is not ideal. Therefore, this paper proposes an error-correcting semi-supervised community detection algorithm (ESCD) based 0 on active learning. It can calculate the traditional k-means algorithm step by step, and adding pairs of constraints in 2 : the clustering process. In order to preserv