Sentiment Analysis

Sentiment analysis (or 'opinion mining') is the field of computational linguistics that deals with the automatic identification of users' emotions, evaluations, and subjective judgements in text. Due to the high diversity of the ways in which people can express their opinions, sentiment analysis is commonly considered to be a highly domain-dependent task. However, lexicon-based approaches can capture the domain-neutral sentiment words to a large extent.

Opinions can be analyzed at different levels of language granularity: for a complete text, a sentence, and on the subsentential level. Finally, a common distinction is between coarse-grained (determine just the polarity or 'semantic orientation' of a piece of text) and fine-grained (determine holder, target, and polarity of the opinion) analysis.

In our group, we work on both English and German and on a range of different domains and genres, including newspaper articles, product reviews and Twitter microblogs. The English work is a collaboration with the group of Prof. Maite Taboada (Simon Fraser Univ., Vancouver), who developed the SO-CAL system (Taboada et al. 2011), and we showed how it can be successfully merged with a document's genre structure analysis (Taboada et al. 2009).

For German, we currently focus on sentiment analysis for Twitter (Sidarenka 2016) and on methods for generation lexical resources for that task (Sidarenka/Stede 2016).

Related Projects

Related publications: