User Mood Tracking for Opinion Analysis on Twitter

摘要

The huge variability of trends, community interests and jargon is a crucial challenge for the application of language technologies to Social Media analysis. Models, such as grammars and lexicons, are exposed to rapid obsolescence, due to the speed at which topics as well as slogans change during time. In Sentiment Analysis, several works dynamically acquire the so-called opinionated lexicons. These are dictionaries where information regarding subjectivity aspects of individual words are described. This paper proposes an architecture for dynamic sentiment analysis over Twitter, combining structured learning and lexicon acquisition. Evidence about the beneficial effects of a dynamic architecture is reported through large scale tests over Twitter streams in Italian.