Sentiment analysis on Italian tweets

International audienceWe describe TWITA, the first corpus of Italiantweets, which is created via a completely automatic procedure, portable to any other language. We experiment with sentiment analysis on two datasets from TWITA: a genericcollection and a topic-specific collection. Theonly resource we use is a polarity lexicon,which we obtain by automatically matchingthree existing resources thereby creating thefirst polarity database for Italian. We observethat albeit shallow, our simple system capturespolarity distinctions matching reasonably wellthe classification done by human judges, withdifferences in performance across polarity values and on the two sets