Learning sentiment classification model from labeled features

摘要: We propose a novel framework where an initial classifier is learned by incorporating prior information extracted from existing sentiment lexicon. Preferences on expectations of labels those lexicon words are expressed using generalized expectation criteria. Documents classified with high confidence then used as pseudo-labeled examples for automatical domain-specific feature acquisition. The word-class distributions such self-learned features estimated the and to train another constraining model's predictions unlabeled instances. Experiments both movie review data multi-domain dataset show that our approach attains comparable or better performance than exiting weakly-supervised classification methods despite no labeled documents.