Reranking Models in Fine-grained Opinion Analysis

Proceedings of the 23rd International Conference on Computational Linguistics (Coling 2010), pages 519–527, Beijing, August 2010 Reranking Models in Fine-grained Opinion Analysis Richard Johansson and Alessandro Moschitti University of Trento {johansson, moschitti}@disi.unitn.it Abstract We describe the implementation of reranking models for fine-grained opinion analysis – marking up opinion expres- sions and extracting opinion holders. The toreranking approach makes it possible model complex relations between multiple opinions in a sentence, allowing us to represent how opinions interact through the syntactic and semantic structure. We carried out evaluations on the MPQA corpus, and the experiments showed significant improvements over a conventional system that only uses local information: for both tasks, our system saw recall boosts of over 10 points. 1 Introduction Recent years have seen a surge of interest in the automatic processing of subjective language. The technologies emerging from this research have ob- vious practical uses, either as stand-alone appli- cations or supporting other NLP tools such as information retrieval or question answering sys- tems. While early efforts in subjectivity analysis focused on coarse-grained tasks such as retriev- ing the subjective documents from a collection, most recent work on this topic has focused on fine- grained tasks such as determining the attitude of a particular person on a particular topic. The devel- opment and evaluation of such systems has been made possible by the release of manually anno- tated resources using fairly fine-grained represen- tations to describe the structure of subjectivity in language, for instance the MPQA corpus (Wiebe et al., 2005). A central task in the automatic analysis of sub- jective language is the indentification of subjective expressions: the text pieces that allow us to draw the conclusion that someone has a particular feel- ing about something. This is necessary for fur- ther analysis, such as the determination of opin- ion holder and the polarity of the opinion. The MPQA corpus defines two types of subjective ex- pressions: direct subjective expressions (DSEs), which are explicit mentions of attitude, and ex- pressive subjective elements (ESEs), which signal the attitude of the speaker by the choice of words. The prototypical example of a DSE would be a verb of statement or categorization such as praise or disgust, and the opinion holder would typi- cally be a direct semantic argument of this verb. ESEs, on the other hand, are less easy to cate- gorize syntactically; prototypical examples would include value-expressing adjectives such as beau- tiful and strongly charged words like appease- ment, while the relation between the expression and the opinion holder is typically less clear-cut than for DSEs. In addition to DSEs and ESEs, the MPQA corpus also contains annotation for non- subjective statements, which are referred to as ob- jective speech events (OSEs). Examples (1) and (2) show two sentences from the MPQA corpus where DSEs and ESEs have been manually annotated. (1) He [made such charges]DSE [despite the fact]ESE that women’s political, social and cul- tural participation is [not less than that]ESE of men. (2) [However]ESE, it is becoming [rather …

Content truncated to 500 words