Study On Sentiment Classification With Emotion Knowledge

Recently, with the emergence of the blog, e-commerce, social intercourse sites and microblogging platforms, subjective text information released by the user is undergoing a rather rapid expansion. To automatically analyze the subjective information, the task of sentiment classification has attracted many researchers’attention and achieved a quick development. This task aims to classify the text into some sentimental categories, such as positive vs. negative.Emotion is the individual’s inherent psychological reactions and feelings. For expressing opinion or sentiment, emotion knowledge has some advantages including:(1) smaller size of the invovled emotion keywords;(2) strong sentiment expression;(3) often domain independent. Therefore, this paper attempts to use emotion knowledge to help improve the classification performance of sentiment classification. The main content and creative points include:First, this paper proposes a novel classification method for sentiment classification with emotion words. This method firstly extracts the automatically-labeled samples with high precision based on the emotion words, and then applies the semi-supervised learning approach to learn a classifier for sentiment classification. Experiments demonstrate that the proposed method achieves better results in different domains.Second, this paper proposes a collaborative learning method with both emotion words and opinion words for sentiment classification. Given these two kinds of keywords, a document-word bipartite graph is built, and these keywords are served as labeled points while the documents are regarded as unlabeled points in the graph. Label-propagation algorithm is then applied to propagate the label information of the words to the documents. Finally, the high confident automatically-labeled samples are used as training data for sentiment classification. Experiments show the proposed method achieves much better classification performance in several domains.Third, this paper proposes a novel domain adoption method for sentiment classification with emotion words. To solve the domain adaptation problem in sentiment classification, this method employs some emotion keywords to extract the automatically-labeled samples with high precision from the target domain. Then both the automatically-labeled samples and the real labeled samples are considered as new labeled data set. Finally, all the labeled data and the unlabeled data in the target domain are used to perform sentiment classification with label-propagation algorithm based on bipartite graph. The expreimental results show that the proposed method effectively enhances the learning ability of unlabeled data on the target domain and improves the adpation ability a lot for sentiment classification.