基于加权内容-结构网络和随机游走的社团划分算法

针对传统模块优化社团划分算法仅能利用网络的结构信息，而无法利用同样丰富的内容信息，导致划分精度较低的问题，提出一种结合内容属性并通过给连边加权来全面优化网络拓扑结构的社团划分算法CCSRW（Classification with Content-Structure and Random Walk）.设计利用随机游走理论计算结构节点与内容节点间的相似性关系矩阵，并将结构节点映射到内容属性空间上，最终把社团划分问题转化为多维无监督聚类问题.通过在真实数据集上进行的全面实验分析，展示了相比于传统社团划分算法，本文的算法能更准确的描述网络结构，显著提高划分性能，并有效解决小社团不敏感问题，更适用于大规模复杂信息网络的社团划分.

Community Detection Based on Weighted Content-Structural Network and Random Walks

For the traditional module optimization community partition algorithms can only use the structure information of network,and cannot use the rich content information,leading to low precision problem.A community partition algorithm that is combined with the content attribute and empowers the edge to fully optimize the topology of the network,called CCSRW (Classification with Content-Structure and Random Walk) is proposed.We use random walk theory to calculate the similarity relationship matrix between structure nodes and content nodes,and map structure nodes onto the content attribute space,finally divide the community partition problems into multidimensional unsupervised clustering problems.Comprehensive experimental analysis on the real data sets shows that compared to the traditional community partition algorithms,this algorithm can describe the network structure more accurately,improve the classification performance significantly,and solve the problem that is not sensitive to small community effectively,and it is more suitable for the large-scale complex information network community partition.