The chinese bag-of-opinions method for hot-topic-oriented sentiment analysis on weibo

With the rapid growth of Weibo, sentiment analysis on the hot topics which are spotlighted suddenly, spread rapidly, and influence widely during a short period becomes crucial. However, because of the urgent analysis requirement and diversity of the hot topics, the state-of-the-art supervised methods would fail due to the lack of annotated training data. To address this problem, we first propose a Chinese bag-of-opinions model based on dependency grammar representing Weibo sentences. Then, we calculate sentiment polarity score for every opinion and get a weighted summation sentiment evaluation for each sentence. A confidence value of a sentence’s polarity score is also defined. With it, we can extract sentences with high confidences as annotated data which can guide further analysis. We applied our model with the summation evaluation and semi-supervised methods. Experiments conducted on the NLP&CC 2012 dataset for Chinese sentiment analysis validate the effectiveness of our method.