Measuring praise and criticism: Inference of semantic orientation from association

摘要: The evaluative character of a word is called its semantic orientation. Positive orientation indicates praise (e.g., "honest", "intrepid") and negative criticism "disturbing", "superfluous"). Semantic varies in both direction (positive or negative) degree (mild to strong). An automated system for measuring would have application text classification, filtering, tracking opinions online discussions, analysis survey responses, chat systems (chatbots). This article introduces method inferring the from statistical association with set positive paradigm words. Two instances this approach are evaluated, based on two different measures association: pointwise mutual information (PMI) latent (LSA). experimentally tested 3,596 words (including adjectives, adverbs, nouns, verbs) that been manually labeled (1,614 words) (1,982 words). attains an accuracy 82.8p full test set, but rises above 95p when algorithm allowed abstain classifying mild