Sentiment classification finds various applications in opinion mining, which can help users determine sentiment tendency of texts and information. In this paper, we consider the problem of text orientation analysis. In particular, we propose a two-stage approach by coupling sentiment dictionary and classification methods. In the first stage, we build sentiment dictionary and rules to obtain the texts whose emotional scores are ranked in the top 1/4 and the bottom 1/4. These texts are marked classified for supervising the second stage. In the second stage, we employ the SVM classifier to process the remaining texts. Finally, we combine the two stages to get the orientation analysis results for all the texts. Experimental results demonstrate that, in contrast to using sentiment dictionary and classification method separately, our proposed method achieves higher classification accuracy when an initial training set by manual tagging is unavailable.
Abstract-concept learning carryover effects from the initial training set in pigeons (Columba livia).
