Abstract: A classifier based on Logistic Regression（LR） is employed to identify comment spams which have flooded in Web forums.Comparative study on performances of LR and Support Vector Machine（SVM） is presented.It is introduced that a relevancy coefficient vector space model named cVSM which is used to express comment archives.Some feature extractive methods are discussed，including Information Gain（IG），Mutual Information（MI），χ² statistic（CHI） and Document Frequency（DF）.The experiments show that：The learn time of LR is less than 1/10 of SVM’s.DF and IG have better performances than MI and CHI.To be compared with vector space model，cVSM has improved comment spam cognitive capability of classifier.

摘要： 针对Web社区垃圾信息泛滥的问题，采用基于Logistic回归（LR）的分类器区分合法评论和垃圾评论，并和支持向量机（SVM）的性能对比。提出了相关度向量空间模型cVSM作为评论的文档表示模型，讨论了信息增益IG、互信息MI、χ²统计CHI、文档频率DF等不同特征抽取方法对模型的影响。实验结果表明，LR的训练时间不到SVM的1/10；DF和IG比MI和CHI表现更好；与传统的向量空间模型相比，使用cVSM显著提高垃圾评论识别能力。