摘要
针对目前复杂网络优化聚类算法目标函数的有偏性影响聚类精度的问题,提出了"群"的概念,实现了对节点在聚类过程中局部信息决策环境的划定。提出了基于"群"概念改进的网络模块性评价函数,并以该函数作为目标函数对Fast-Newman(FN)算法进行了改进。在不同类别数据集上进行的聚类实验的结果表明,基于"群"思想改进的FN算法(GFN)在复杂网络中的聚类精度比FN算法平均提高了约70%,从而验证了"群"思想在揭示真实簇结构过程中的有效性。
To deal with the problem that the object function of existing optimized clustering algorithms are biased, which may affect the accuracy of the clustering, the concept of groups was proposed in this paper, to model the local con- text of nodes during the clustering process. An improved modularity function based on the concept of groups was giv- en, and the GFN, a clustering algorithm derived from the well-known Fast-Newman algorithm. Experiments on differ- ent datasets showed that the new method increased the clustering accuracy by 70% on average compared with the original version, proving that the group concept is significant in depicting the actual clustering structures in real net- works.
出处
《高技术通讯》
CAS
CSCD
北大核心
2013年第10期1016-1023,共8页
Chinese High Technology Letters
基金
国家自然科学基金(60873241
61170296
61190120)
软件开发环境国家重点实验室基金(SKLSDE-2012ZX-17)
航空科学基金(20091951020)
新世纪优秀人才支持计划(NECT-09-0028)
北京自然科学基金(4123101)资助项目
