Partially Supervised Phrase-Level Sentiment Classification

Abstract

This paper presents a new partially supervised approach to phrase-level sentiment analysis that first automatically constructs a polarity-tagged corpus and then learns sequential sentiment tag from the corpus. This approach uses only sentiment sentences which are readily available on the Internet and does not use a polarity-tagged corpus which is hard to construct manually. With this approach, the system is able to automatically classify phrase-level sentiment. The result shows that a system can learn sentiment expressions without a polarity-tagged corpus.