Web论坛上的垃圾贴过滤

随着网络的发展,Web论坛成为Web用户信息共享和分组合作的新平台.Web论坛上积累了海量的知识,由此成为互联网上进行数据挖掘的宝贵资源.在Web论坛上的应用常受到论坛上低质量帖子(垃圾贴)的影响.因此针对在Web论坛上进行垃圾贴过滤的问题,提出了基于隐含狄利克雷分布的CJTM和CAJTM模型,CJTM和CAJTM模型利用了论坛帖子的文本内容,帖子间的回复链接信息和作者信息,和传统的分类方法及基于规则的方法相比,CJTM和CAJTM模型不需要训练集和规则集.在实际Web论坛数据中进行的实验显示出较好的效果.

Junk Post Filtering in Web Forums

Web forums have emerged to Web users as new platforms for information sharing and group collaboration.With the large volume of accumulated knowledge,Web forums have become valuable sources for data mining in recent years.However,the performance of those data mining applications is usually harmed by low-quality-posts in Web forums.In this paper,the problem of filtering lowquality-posts(junk posts)in Web forums is focused on.Inspired by LDA,graphic models,clustered junk topic model and clustered author junk topic model,are built for detecting junk posts.Text contents,reply linkages and author information are utilized in the presented models.In contrast to traditional approaches such as classification methods,these models require no training and no hard coded rule sets.The presented models can help to not only understand the process of generating posts but also quantitatively evaluate the quality of post contents in Web forums.Experiments conducted on a real Web forum show that this approach achieves better results compared with traditional methods.