Enhancing Opinion Extraction by Automatically Annotated Lexical Resources (Extended Version)

In this paper we tackle an opinion extraction (OE) task, i.e., identifying in a text each expression of subjectivity, the subject expressing it, and its possible target. We especially focus on how lexical resources specifically developed for opinion mining could be used to improve the performance of an opinion extraction system. We report results, complete with statistical significance tests and inter-annotator agreement data, on two manually annotated corpora, one of English and one of Italian texts. We evaluate our results using standard evaluation measures and also using a new evaluation measure we have recently proposed.