Sentiment analysis Aim

Automatically detect in written texts whether a sentence/document is positive, negative or neutral (Liu, 2012). Dual perspective: ◦ “When the usage of a word gives an impression of an attitudinal or pragmatic meaning, this is called a semantic prosody ” (Sinclair, 1999) ◦ "Polarity identification focuses on whether a language unit has a positive or negative connotation (Su and Markert, 2008) Applications • Products (especially movies and books): is this review positive or negative? • Organisations/Politicians: what do people think about them? • Services/Events: how is consumer confidence? • News: what are the author’s opinions? • Politics: what do people think about this issue? • Prediction: predict election outcomes or market trends 3 Conventional approach No word order = sum of single sentiments to get the overall one (Pang et al, 2002). E.g.: “I will eliminate capital gains taxes for the small businesses” Capital + businesses + I + gains + will + small + the + taxes + for + eliminate = 4 Issues Polarity reversals eliminate taxes Negations Our government should work for us, not against us Different targets and sentiments Hundreds of billions in tax breaks for big corporations, but not one penny of tax relief to Americans Pronominal references I don't believe that Senator McCain doesn't care. I just think he doesn't know. False sentiment words Well, it's time for them to own their failure Opinions implied Today, we import triple the amount of oil than we had on the day that Senator McCain took office Idioms, Metaphors and Sarcasm Pull yourself up by your own bootstraps, even if you don't have boots 5 Pairs consisting of one target and one modifier (I will eliminate) + (capital gains taxes) + (for the small businesses) = Word meaning and sentence structure important to calculate the overall sentiment. Proposed approac