基于论文相似网络拓扑结构的聚类方法比较

以R语言中的复杂网络处理包igraph为工具，基于语义相似性算法构建论文相似网络，然后采用四种代表性网络聚类算法（随机游走法、标签传播法、最大模块度法、边介数法）对构建出的网络进行聚类分析。最后结合金标准和网络社团划分评价指标D函数比较四种算法的准确性和稳定性，发现随机游走算法最为卓越，同时明确了复杂网络的预处理也是一个影响聚类效果的重要因素。

Comparison of clustering methods in light of paper similarity network topology

The complex network processing package igraph in R was applied to clustering analysis paper similarity network built by semantic similarity algorithms with four representative algorithms (Walktrap algorithm , Label propagation algorithm , BGll algorithm, Girvan-Newman algorithm ). Finally, we compared the stability and precision of clustering algorithms using gold standard and a D function of network community detection index. Found that the walktrap algorithm was the most outstanding, at the same time, the pre-processing of complex networks was also an important factor which can influence clustering effect.