Integrating intra- and inter-document evidences for improving sentence sentiment classification

Sentence sentiment classification is an important task of sentiment analysis. It aims to classify the sentences into positive, negative, or objective. One can consider sentence sentiment classification as a standard text categorization problem. However, determining the sentiment orientation of a review sentence requires more than the features inside the sentence itself, especially for the sentences with little or ambiguous inside sentence features. Through observing, some features outside the sentence can interact with its inside features to enhance the overall performance of sentence sentiment classification. Thus in this paper, we propose two such outside sentence features: intra-document evidence and inter-document evidence. Then in order to improve the sentence sentiment classification performance, a graph-based propagation approach is presented to incorporate these inside and outside sentence features. The experimental results on camera domain show that the proposed approach performs better than the approaches without using outside sentence features, and outperforms other representational previous approaches.