News sentiment analysis based on cross-domain sentiment word lists and content classifiers

The main task of sentiment classification is to automatically judge sentiment polarity (positive or negative) of published sentiment data (e.g. news or reviews). Some researches have shown that supervised methods can achieve good performance for blogs or reviews. However, the polarity of a news report is hard to judge. Web news reports are different from other web documents. The sentiment features in news are less than the features in other Web documents. Besides, the same words in different domains have different polarity. So we propose a self-growth algorithm to generate a cross-domain sentiment word list, which is used in sentiment classification of Web news. This paper considers some previously undescribed features for automatically classifying Web news, examines the effectiveness of these techniques in isolation and when aggregated using classification algorithms, and also validates the self-growth algorithm for the cross-domain word list. ? Springer-Verlag 2012.EI