Knowledge discovery from online social network

个人之间以电子格式(如电子邮件、即时消息、博客等)进行的通信量不断增加推动了社交网络分析中的计算研究。社交网络分析技术旨在搜索具有共同兴趣的社区或社区内的领导人。社交网络通常被表示为图，其中节点代表个体，边代表他们之间的关系。这样的图是海量的，其中的节点可能包含大量的文本数据。许多现有的社交网络分析技术要么关注通过通信频率测量的社交网络拓扑，要么关注用户生成的内容。然而，这两个信息本身都不足以准确地找到具有共同利益的社区和社区内的领导人。文本中的信息和链接结构相互促进，从而产生更高质量的结果。除此之外，现有的社交网络分析技术只能有效地分析来自单一来源且相对完整的图表。此外，现有的大多数方法都假设网络结构是静态的。然而，在线社交网络不断变化，网络中的链接来自不同的在线来源。例如，考虑从Usenet到博客圈的链接、推文和新闻文章之间的链接等。还有一些应用程序无法同时访问整个网络，但以连续流的形式提供。这样的应用程序带来了独特的挑战，因为整个图形不能保存在主内存中。要使社交网络分析更有效，需要开发考虑文本内容、不确定性、不完备性、数据源的异构性以及为涉及连续边缘流的Web应用开发专门算法的技术。我们的目标是通过开发适当的模型和算法来有效地挖掘在线社交网络来解决这些问题。

The increasing amount of communication between individuals in e-formats (e.g. email, instant messaging, blogs, etc.) has motivated computational research in social network analysis. Social network analysis techniques aim to search communities of shared interests or leaders within communities. Social networks are often represented as graphs, where nodes represent individuals and edges represent the relationship between them. Such graphs are massive, in which node may contain a large amount of text data. Many existing social network analysis techniques focus either on the social network topology measured by communication frequencies or the content generated by the users. However, neither information alone is sufficient for finding accurately communities of shared interests and leaders within communities. The information in the text and the linkage structure re-enforce each other, and this leads to higher quality result. In addition to this, existing social network analysis techniques are only effective in analyzing graphs which are from a single source and relatively complete. Furthermore, most existing approaches assume that the structure of the network is static. However, online social networks change continually and links within the network come from different online sources. For example, consider links from Usenet to the blogosphere, links between tweets and news articles, etc. There are also some applications in which the whole network is not available at one time, but available in the form of continuous stream. Such applications create unique challenges, because the entire graph cannot be held in main memory. What is needed to make social network analysis more effective is to develop techniques that take into account textual content, uncertainty, incompleteness, heterogeneity of data sources and the need of developing specialized algorithms for Web applications that involve continuous stream of edges. Our goal is to address these issues by developing appropriate models and algorithms for mining effectively online social networks.