有向网络下的CoDA社区发现算法评估

Evaluation of the CoDA community detection algorithm based on directed network

CoDA算法是一种基于概率模型的能识别二分结构的社区发现算法。为了验证该算法的社区划分效果,采用信息检索领域的F-measure标准,对有向网络下重叠社区和非重叠社区的CoDA社区发现算法进行评估。F-measure标准中F1-measure值的大小能反映CoDA算法社区划分效果的优劣。实验所用的数据集由LFR Benchmark工具生成,数据集中节点数最小为100,最大为20 000,每增加100节点对CoDA算法社区划分效果评估一次。分析实验结果可以得出,当节点数小于1 600时,CoDA算法的划分效果较好。当节点数大于1 600时,随着节点个数增多,CoDA算法社区划分效果逐渐变差。由此说明,基于概率模型的CoDA算法适用于小规模社交网络社区的划分。

CoDA（Communities through Directed Affiliations）algorithm is a kind of community detection algorithm which can successfully detect 2-mode communities based on probability model.The F-measure criterion,for information retrieval,is adapted to the evaluation of CoDA algorithm in directed networks with overlapping communities or non-overlapping communities.The value of F1-measure in F-measure criterion can reflect whether CoDA algorithm performs well or not.The data sets used in the experiment is generated by the LFR Benchmark tool.The minimum number of nodes in data set is 100 and the maximum is 20 000,and evaluated experiment is conducted when every 100 nodes is added.The results show that CoDA algorithm performs well when the number of nodes is bellow 1 600.CoDA algorithm＇s performance becomes worse with the increase of the number of nodes,which proves the CoDA algorithm based on probability model is applicable to the community detection of small-scale networks.