基于不平衡情感分类的Lasso-Lars特征选择方法研究

基于Lasso回归和支持向量机分类器,首先利用Lasso回归具有变量筛选的特点,过滤部分不重要的特征,然后利用支持向量机分类器做情感提取.在某化妆品品牌的评论数据实验中,利用基础情感词典和领域情感词典构建待选择高维特征集,通过对比特征选择前后的G-means,精确度和召回率等,均取得显著效果.

Feature Selection in Imbalanced Sentiment Classification: A Method Using Lasso-Lars

The characteristics of textual emotion analysis are usually of high dimension and sparseness. Lasso has a simple and efficient trait in feature selection. This paper introduces the Lasso regression into the unbalanced emotion analysis and achieves remarkable results. Applying emotional analysis in e-commerce plays an important role in improving product quality and improving service, which attracts many researchers and has high research value. In fact, the number of positive comments on e-commerce data generally exceeds the number of bad reviews. If the feature selection is not reasonable, it is easy to ignore the bad reviews, and the bad reviews are the key to analyzing the problems. Based on the Lasso regression and SVM classifier, this paper first uses Lasso regression to filter the features that have variable screening, filters some unimportant features, and then makes use of SVM classifier to extract the emotion. In a cosmetic brand''s reviewing data experiment, the basic emotion dictionary and domain sentiment lexicon are used to construct the high-dimensional feature set to be selected, and the significant effects are achieved by comparing G-means before and after feature selection, accuracy and recall.