就向量空间模型文本表示方法以及归一化技术对不良文本过滤性能的影响进行了研究,并基于平衡样本集和不平衡样本集分别进行了试验。试验和结果分析表明,Naive Bayes方法由于采用概率模型进行文本表示,在不平衡样本集上显示了较差的准确度,而基于向量空间模型进行文本表示的方法,如中心向最法(VSM)、支持向量机(SVM)等在平衡或非平衡样本上取得了较好的准确度,并用于过滤不良文本的文本内容安全监管中。
This paper researches the vector space model for expressing text, and two datasets are used to evaluate the text expressing method, one is a balance data set, the other is a non-balance data set, which is used for filtering some specific text. It gets good precision using VSM and SVM on both data sets, however the result is poor using Naive Bayes model on the non-balance data set, especially to filter unseen reactionary Web text. The paper concludes that term weighting and normalization are very important technique to improve the precision.