基于语义相似度和标签预分配的重叠社区发现方法

Overlapping community detection algorithm based on semantic similarity and label pre-allocation

为解决标签传播算法(COPRA)在划分社区时初始标签多、标签更新随机性强及传播未考虑社交网络中语义的问题,提出基于语义相似度和标签预分配的重叠社区发现方法(OCDSLP)改进社区划分过程,提高社区发现质量。首先,基于Word2Vec模型对用户的博文进行建模,度量用户语义相似度。接着,利用网络拓扑结构特征排序节点并融合语义信息进行标签预分配,减少初始标签数量。最后,在选择标签时融入节点相似性,限制社区的规模,发现语义一致的社区。实验结果表明,OCDSLP可以显著提高发现社区质量。

In order to solve the problems of the label propagation algorithm(COPRA)when dividing communities,the initial label is multiple,the label update is random,and the propagation does not consider the semantics in social networks,we proposed a method OCDSLP to improve the community detection based on semantic similarity and label pre‑allocation.Firstly,based on the Word2Vec model,the user’s blog post is modeled to measure the user’s semantic similarity.Then,the network topology feature or‑dering node and the semantic information are used to pre‑allocate tags to reduce the initial number of tags.Finally,incorporate node similarity when choosing labels,limit the size of the community,and discover communities with consistent semantics.Experi‑mental results have shown that OCDSLP can significantly improve the quality of discovery communities.