Recognizing Contextual Polarity in Phrase-Level Sentiment Analysis

The paper introduces a new approach to phrase-level sentiment analysis built around a two-stage pipeline. The system first determines whether a given expression is neutral or polar, and then, for those judged polar, disambiguates the specific polarity of the expression. This staged design lets the system focus its polarity disambiguation only on expressions that actually carry sentiment, and it targets contextual polarity rather than assuming a fixed prior polarity for each word or phrase.

With this approach, the system is able to automatically identify the contextual polarity for a large subset of sentiment expressions, achieving results that are significantly better than baseline. The work was influential because it emphasized that the polarity of a sentiment expression depends on its context rather than being fixed, and the associated resources for subjectivity and phrase-level sentiment became widely used in later sentiment analysis research.