This paper introduces a novel approach to phrase-level sentiment analysis, focusing on determining whether an expression is neutral or polar and then disambiguating the polarity of polar expressions. The system uses a two-step process: first, it classifies each phrase containing a clue as neutral or polar; second, it disambiguates the polarity of polar phrases. This method significantly improves the accuracy of identifying contextual polarity for a large subset of sentiment expressions compared to baseline methods. The authors also describe a manual annotation scheme for contextual polarity and conduct an inter-annotator agreement study, achieving 82% agreement. The paper includes a detailed analysis of the features used in the classification process and compares the performance of different classifiers. Finally, it discusses related work and concludes by highlighting the effectiveness of the proposed approach.This paper introduces a novel approach to phrase-level sentiment analysis, focusing on determining whether an expression is neutral or polar and then disambiguating the polarity of polar expressions. The system uses a two-step process: first, it classifies each phrase containing a clue as neutral or polar; second, it disambiguates the polarity of polar phrases. This method significantly improves the accuracy of identifying contextual polarity for a large subset of sentiment expressions compared to baseline methods. The authors also describe a manual annotation scheme for contextual polarity and conduct an inter-annotator agreement study, achieving 82% agreement. The paper includes a detailed analysis of the features used in the classification process and compares the performance of different classifiers. Finally, it discusses related work and concludes by highlighting the effectiveness of the proposed approach.