Combining  Link and Content for Community Detection: A Discriminative Approach

综合链接和内容的社区发现：一个歧视性的方法

In  this paper, we consider the problem of combining link and content analysis  for community detection from networked data, such as paper citation networks  and Word Wide Web. Most existing approaches combine link and content  information by a generative model that generates both links and contents via  a shared set of community memberships. These generative models have some  shortcomings in that they failed to consider additional factors that could  affect the community memberships and isolate the contents that are irrelevant  to community memberships. To explicitly address these shortcomings, we  propose a discriminative model for combining the link and content analysis  for community detection. First, we propose a conditional model for link analysis  and in the model, we introduce hidden variables to explicitly model the  popularity of nodes. Second, to alleviate the impact of irrelevant content  attributes, we develop a discriminative model for content analysis. These two  models are unified seamlessly via the community memberships. We present  efficient algorithms to solve the related optimization problems based on  bound optimization and alternating projection.

Extensive  experiments with benchmark data sets show that the proposed framework  significantly outperforms the state-of-the-art approaches for combining link  and content analysis for community detection.

本文中，我们考虑了将链接和内容分析结合起来从网络数据中探测社区的方法，例如文献引用网络以及WWW。   大部分现存的方法通过一个产生式的模型经由社区用户关系产生共同的集合来结合链接和内容信息。这些产生式的模型有一些缺陷，他们不能考虑额外的可能会影响社区关系的因素，而且孤立了和社区关系的不相关的内容。为了清楚地解决这些短板，我们提出了一个歧视性的模型来将链接和内容分析结合起来做社区探测。首先，我们提出了一个条件模型来在模型中进行链接分析。我们引入了一个隐变量来明确地模拟节点的流行度。然后，为了减轻不相关内容的影响力，我们建立了一个歧视性模型来进行内容分析。这两个模型通过社区关系无缝连接。我们提出了有效的算法基于弹回优化和轮流投影来解决相关优化问题。大量在标准测试数据集上的实验证明我们的方法比最好的方法要好，特别是在社区探测的链接组合和内容分析上。

本文运用的是经典的KDD文章的写法。首先讲述背景，提出问题。这个问题是由于别人没有特别注意到。然后分析问题的challenges。这个问题的本质就是将链接分析和内容分析结合起来。最后是在标准数据集上去探测分析。