The paper introduces a method for inferring the semantic orientation (positive or negative) of a word from its statistical association with a set of positive and negative paradigm words. The method, called Semantic Orientation from Association (SO-A), is evaluated using two different measures of word association: Pointwise Mutual Information (PMI) and Latent Semantic Analysis (LSA). The experiments use 3,596 words (1,614 positive and 1,982 negative) from the General Inquirer lexicon. The accuracy of the method is 82.8% on the full test set, but rises above 95% when the algorithm is allowed to abstain from classifying mild words. The paper also discusses related work, including supervised learning algorithms and applications such as text classification, review classification, and sentiment analysis.The paper introduces a method for inferring the semantic orientation (positive or negative) of a word from its statistical association with a set of positive and negative paradigm words. The method, called Semantic Orientation from Association (SO-A), is evaluated using two different measures of word association: Pointwise Mutual Information (PMI) and Latent Semantic Analysis (LSA). The experiments use 3,596 words (1,614 positive and 1,982 negative) from the General Inquirer lexicon. The accuracy of the method is 82.8% on the full test set, but rises above 95% when the algorithm is allowed to abstain from classifying mild words. The paper also discusses related work, including supervised learning algorithms and applications such as text classification, review classification, and sentiment analysis.