Exploring the use of linguistic features in sentiment analysis

Abstract

In this paper we describe some explorations of the potential of genre-revealing features on automatic sentiment analysis. In particular, we use a small subset of the ‘linguistic facets’ employed in recent experiments on automatic genre identification in combination with more traditional sentiment-revealing features on two different single-genre corpora: a corpus of English blogs and a corpus of French reviews(relectures). Although still preliminary, results show that linguistic facets might have a positive influence on sentiment analysis because 6 out of 14 facets used in the experiments are among the first 22 most important discriminative features.