摘要
针对近邻传播聚类算法偏向参数难选定、生成的簇数目偏多等问题,提出一种概率无向图模型的近邻传播聚类算法.首先为样本数据构建概率无向图模型,利用极大团和势函数计算无向图中数据样本的概率密度,将此概率密度作为一种聚类先验知识注入近邻传播算法的偏向参数中,提高算法的聚类效率;并用高斯降噪和簇归并方法进一步提升算法的聚类精度.在UCI数据集上的实验结果表明,所提出算法的聚类效率和精度均优于相比较的同类算法.
In order to solve the problem that the preference of the traditional affinity propagation clustering algorithm is difficult to choose and the number of generate clusters is likely to be overmuch, an affinity propagation clustering method based on the probabilistic undirected graph model is proposed in this paper. Firstly, the probabilistic undirected graph model is constructed for sample data, while the probability density is calculated for each sample data by maximum clique and potential function. Then the probability density as a priori clustering knowledge is put into the preference of the affinity propagation algorithm to improve its efficiency. The clustering accuracy of the algorithm is further promoted by using the Gauss noise reduction and cluster merging method. Experimental results on the UCI data sets show better clustering efficiency and accuracy of the proposed algorithm against several other similar algorithms.
作者
覃华
詹娟娟
苏一丹
QIN Hua ZHAN Juan-juan SU Yi-dan(College of Computer and Electronic Information, Guangxi University, Nanning 530004, Chin)
出处
《控制与决策》
EI
CSCD
北大核心
2017年第10期1796-1802,共7页
Control and Decision
基金
国家自然科学基金项目(61363027)
教育部人文社会科学研究规划基金项目(11YJAZH080)
关键词
近邻传播聚类算法
偏向参数
概率无向图模型
高斯平滑
簇归并
affinity propagation clustering algorithm
preference
probabilistic undirected graphical model
gaussian smooth
cluster merging
