基于LDA的结构-内容联合社团发现模型

在社会网络中，根据已有的连接关系和文本信息发掘社会网络中的社团不但可以将相似的用户划分在一个社团，还可以用来预测网络中潜在的连接关系。为了提高社会网络中社团发现的性能，本文提出了一种基于LDA的结构-内容联合社团发现模型。首先，对社会网络的图论描述进行转化，使其适用于LDA模型。其次，对LDA模型描述进行扩充，使其包含了用户间交互的文本信息。最后，通过Gibbs采样方法对模型的参数进行估计。实验表明，本文提出的社团发现模型与其它相关方法相比较，社团发现得到的社团不仅用户间连接的紧密度和用户共享兴趣爱好的强度高，而且可以更好地用于社会网络中潜在连接的预测。

LDA Based Community Discovery Model Jointly Considering Structure and Content

Community discovery is one of the most important issues in social networks. In social networks, discovering communities according to the structure and content of networks can not only group users in communities, but also predict latent relationship between users. In order to improve the performance of community discovery, this paper proposed a LDA based community discovery model jointly considering structure and content. Firstly, we transformed the graph of a social network into a LDA based model. Secondly, we extended the transformed LDA model by incorporating contents of actions between users. Finally, we estimated the parameters of the proposed model based on Gibbs sampling. The experiments show that, compared with related works, the proposed model is more efficient in discovering communities among which users are tightly connected and having similar interests. Moreover, based on the communities discovered, we can better predict the latent friendships in a social network.