PolArt: a robust tool for sentiment analysis

Abstract

We introduce PolArt, a robust tool for sentiment analysis. PolArt is a pattern-based approach designed to cope with polarity composition. In order to determine the polarity of larger text units, a cascade of rewrite operations is carried out: word polarities are combined to NP, VP and sentence polarities. Moreover, PolArt is able to cope with the target-specific polarity of phrases, where two neutral words combine to a non-neutral phrase. Target detection is done with the Wikipedia category system, but also user defined target hierarchies are allowed. PolArt is based on the TreeTagger chunker output, and is customised for English and German. In this paper we evaluate PolArt’s compositional capacity.