Extraction of multiple communities based on information diffusion results on a large social network

在这项工作中，我们从Twitter上的一个大型社交网络中提取了多个相互重叠、但对不同事物感兴趣的社区。Twitter是一家著名的微博服务公司。为此，我们首先通过文章中的标签和特征关键词以及一种主题估计方法(LDA：潜在狄利克雷分配)的结果来提取信息扩散网络。然后，通过图挖掘技术将这些网络整合在一起，该技术可以从多个图中发现频繁模式。也就是说，如果信息扩散网络共享频率等于或大于给定阈值的公共子结构，则它们是集成的。此外，我们设计了一种准确检测信息扩散序列中扩散速度发生变化的变化点的方法，以研究结果社区的特征。

In this work, we extracted multiple communities that overlap with each other, but have interest in different things from a large social network on Twitter, a notable microblogging service. To this end, we first extracted information diffusion networks by means of tags and characteristic keywords in articles, as well as results of a topic estimation method (LDA: Latent Dirichlet Allocation). Then, those networks are integrated by means of a graph mining technique that can find frequent patterns from multiple graphs. Namely, information diffusion networks are integrated if they share common substructures whose frequency is equal to or greater than a given threshold. Furthermore, we devised a method of accurately detecting change points in information diffusion sequences around which diffusion speed has changed in order to investigate the features of resulting communities.