This paper presents a holistic lexicon-based approach to opinion mining, focusing on determining the semantic orientations (positive, negative, or neutral) of opinions expressed on product features in customer reviews. The approach addresses two major challenges: handling context-dependent opinion words and aggregating multiple conflicting opinion words in a sentence. The system, called Opinion Observer, uses external evidence and linguistic conventions to infer the semantic orientation of opinion words, without relying on prior domain knowledge. The proposed technique is evaluated using a benchmark dataset and additional reviews, showing significant improvements over existing methods in terms of precision and recall. The results demonstrate the effectiveness of the proposed approach in accurately identifying the semantic orientations of opinions, making it a valuable tool for opinion mining and related applications.This paper presents a holistic lexicon-based approach to opinion mining, focusing on determining the semantic orientations (positive, negative, or neutral) of opinions expressed on product features in customer reviews. The approach addresses two major challenges: handling context-dependent opinion words and aggregating multiple conflicting opinion words in a sentence. The system, called Opinion Observer, uses external evidence and linguistic conventions to infer the semantic orientation of opinion words, without relying on prior domain knowledge. The proposed technique is evaluated using a benchmark dataset and additional reviews, showing significant improvements over existing methods in terms of precision and recall. The results demonstrate the effectiveness of the proposed approach in accurately identifying the semantic orientations of opinions, making it a valuable tool for opinion mining and related applications.