LAUSR

As a fundamental task of sentiment analysis, document level sentiment classification aims to predict user’s overall sentiment (e.g., positive or negative) towards the target in a document. The document usually consists of various opinion sentences towards different aspects with different sentiments. Therefore, the overall opinion towards the whole target should play a more important role in document sentiment prediction. However, most existing methods for the task treat all sentences of the document equally. Thus, they are easy to encounter difficulty when the sentiments of most aspect opinion sentences are not coherent with the overall sentiment. To address this, we propose a novel method for document sentiment classification which adequately explores the effect of overall opinion sentences. In our method, firstly, multiple features are exploited to recognize candidate overall opinion sentences, and then a structural SVM is utilized to encode the overall opinion sentences for document sentiment classification. Experiments on several public available datasets including product reviews and movie reviews show the effectiveness of our method.