基于注意力网络特征的社区发现算法

现实世界中许多网络都是根据社区结构紧密组织起来的, 发现社区对于了解复杂网络的结构及其关系有很大的帮助, 文中提出了一种基于注意力网络特征的社区发现(community discovery algorithm based on attention network features, CANF)算法, 利用标记节点频率和反示例节点频率度量初始网络标记特征, 并且引入注意力机制, 对示例节点的每个邻居节点更好地分配权重, 将初始权重与分配权重相结合, 使初始度量的网络特征获取更多与目标有关的细节信息。文中通过分配的注意力网络特征进行复杂网络预处理以及社区博弈归并, 于真实网络中进行验证, 实验结果表明, CANF算法在准确度、模块度以及运行时间方面优于其他社区发现算法。

Community discovery algorithm based on attention network feature

In the real world, many networks are closely organized according to the community structure. It has been found that the community is of great help in understanding the structure and relationship of complex networks. This paper proposes a community discovery algorithm based on attention network features (CANF algorithm). It uses the label node frequency and the inverse example node frequency to measure the initial network label characteristics, and introduces the attention mechanism, and updates each neighbor node of the example node. Combined the initial weight with the assigned weight, the network features of the initial measure can obtain more detailed information related to the target. In this paper, complex network preprocessing and community game merge are carried out through distributed attention network features. The experimental results show that the accuracy, modularity and running time of the community discovery algorithm based on attention network features are better than other community discovery algorithms.