Abstract: Currently, methods for community analysis can’t make full use of the intrinsic properties of complex systems, so it is difficult to find a correlation among groups in complex weighted network accurately and effectively. Based on the active property of stock network, this paper proposes a hierarchical community detection algorithm for composite weighted stock network. This algorithm makes the definition on stock activity, puts forward a composite weighted model to effectively represent the activity of stock network, and presents the criteria to evaluate the dissimilarity among groups in order to achieve the community detection for composite weighted network. This algorithm constructs a composite weighted stock network on correlation for the side of the stock price volatility. Taking the turnover and volume of stock as evaluation standard, it chooses high-activeness stocks. Furthermore, centered on the chosen stocks, based on stock inter-dissimilarity weight price guidelines, it extracts multiple high active local structures, thus effectively avoids the matter that the community can’t be discovered in right direction because of the low quality initial community structure. Finally, based on the high active local structures, this paper divides the community in weighted network by globally optimizating the modularity. The CNM algorithm and BGLL algorithm are applied to the established network, the comparison results show the superiority of the algorithm.

Key words: activeness, composite weighted, community detection, turnover, volume

摘要： 当前社团分析方法没有充分利用复杂系统的内在特性，难以准确和有效地发现复杂加权网络群体之间的相关性。基于股票网络的活跃性，提出了一种基于活跃性的复合加权股票网络的层次社团划分算法。该算法对股票活跃性进行了定义，提出了一种复合加权模型以有效表示股票网络的活跃性，进而为了实现复合加权网络的社团划分，给出了群体相异度的评判标准。该算法以股票价格波动的相关性为边建立复合加权股票网络，以股票的换手率和成交量为评价标准，选出活跃性高的股票，进而以活跃性股票为中心，基于股票间的相异度权重标价准则，提取多个高活跃的局部结构，可以有效避免层次划分算法由于初始社团结构质量不高，导致社区结构不能沿正确方向继续进行层次发现的问题。最后，基于高活跃的局部结构性，利用全局优化模块度的方法对复合加权网络进行社团划分。将CNM算法（Newman贪婪算法）与BGLL算法运用于构建的网络中，结果表明了算法的优越性。

关键词: 活跃性, 复合加权, 社团划分, 换手率, 成交量