垃圾评论自动过滤方法  

Method of review spam detection

作　　者：谭文堂[1] 朱洪[1] 葛斌[1] 李芳芳[1] 肖卫东[1] 

机构地区：[1]国防科技大学信息系统工程重点实验室,湖南长沙410073

出　　处：《国防科技大学学报》2012年第5期153-157,168,共6页Journal of National University of Defense Technology

基　　金：国家自然科学基金资助项目(60903225);国防科技大学优秀研究生创新基金资助项目(S100502)

摘　　要：针对互联网上存在的大量垃圾评论,提出一种基于电阻网络的垃圾评论检测方法,该方法用电阻距离来度量评论之间的上下文语义相似性,把整个评论数据表示成一个电阻网络,把垃圾评论当作该网络上的语义离群点来处理,根据网络节点对电阻网络平均电能消耗的影响,建立电离群因子来度量数据的离群程度,以此来识别垃圾评论。实验证明了该方法的有效性,在多个数据集上取得了较好的效果。For detecting review spam in the Internet automatically,an method based on resistance network is proposed.By treating the distance between two reviews as a resistance,we represent the given dataset as a resistance network and the resistance distance between two nodes is a measure of the semantic distance between them.Spam reviews are semantic outliers in this network.An electrical outlier factor（EOF） for each review based on its influence on the power dissipated of the network was used to detect the spam reviews.Experimental results testified that EOF is suitable for detecting review spam,and is efficient and effective.

关 键 词：垃圾评论检测 电阻距离 电离群因子 

分 类 号：TP316[自动化与计算机技术—计算机软件与理论]