摘要
针对极大熵聚类算法在处理多视角聚类任务时存在的局限性,引入划分融合和视角加权技术,提出一种改进的极大熵聚类算法。通过对视角分配权重体现其重要程度,在此基础上对每个视角进行单独划分,利用融合权重矩阵实现视角划分的融合,并采用新的集成策略得到全局聚类结果。在人工数据集和UCI数据集上的实验结果表明,与极大熵聚类算法、基于多任务的组合K-means算法等相比,该算法具有更好的多视角聚类性能。
Aiming at the limitation to effectively realize the view fusion in the multi-view clustering task for Maximum Entropy Clustering （ MEC）, this paper proposes an improved view-weighting MEC algorithm by introducing partition fusion and view-weighting. This method assigns a weight of each view to show the importance of each view. And it sets the partition matrix of each view, and the view-fusion in each view partition is made by a view-fusion weighting matrix, Finally,it proposes a new integration strategy to obtain the global partition result. Experimental results on synthetic datasets and UCI datasets show that the proposed algorithm outperforms MEC algorithm and CombKM algorithm in dealing with multi-view clustering task.
出处
《计算机工程》
CAS
CSCD
北大核心
2016年第4期184-189,196,共7页
Computer Engineering
基金
国家自然科学基金资助面上项目(61170122)
江苏省杰出青年基金资助项目(BK20140001)
新世纪优秀人才支持计划基金资助项目(NCET120882)
关键词
极大熵聚类
多视角聚类
划分融合
视角加权
权重矩阵
Maximum Entropy Clustering （MEC）
multi-view clustering
partition fusion
view-weighting
weight matrix
