Discourse Level Opinion Interpretation

Proceedings of the 22nd International Conference on Computational Linguistics (Coling 2008), pages 801–808 Manchester, August 2008 Discourse Level Opinion Interpretation ∗ Swapna Somasundaran Dept. of Computer Science University of Pittsburgh Pittsburgh, PA 15260 swapna@cs.pitt.edu Janyce Wiebe Dept. of Computer Science University of Pittsburgh Pittsburgh, PA 15260 wiebe@cs.pitt.edu Josef Ruppenhofer Intelligent Systems Program University of Pittsburgh Pittsburgh, PA 15260 josefr@cs.pitt.edu Abstract This work proposes opinion frames as a representation of discourse-level associa- tions which arise from related opinion top- ics. We illustrate how opinion frames help gather more information and also assist disambiguation. Finally we present the re- sults of our experiments to detect these as- sociations. 1 Introduction Opinions have been investigated at the phrase, sen- tence, and document levels. However, little work has been carried out regarding interpreting opin- ions at the level of the discourse. Consider the following excerpt from a dialog about designing a remote control for a television (the opinion targets – what the opinions are about — are shown in italics). (1) D :: And I thought not too edgy and like a box, more kind of hand-held not as computery, yeah, more or- ganic shape I think. Simple designs, like the last one we just saw, not too many buttons. Speaker D expresses an opinion in favor of a design that is simple and organic in shape, and against an alternative design which is not. Several individual opinions are expressed in this passage. The first is a negative opinion about the design be- ing too edgy and box-like, the next is a positive opinion toward a hand-held design, followed by a negative opinion toward a computery shape, and so on. While recognizing individual expressions ∗This research was supported in part by the Department of Homeland Security under grant N000140710152. ∗c⃝ 2008. Licensed under the Creative Commons Attribution-Noncommercial-Share Alike 3.0 Unported li- cense (http://creativecommons.org/licenses/by-nc-sa/3.0/). Some rights reserved. of opinions and their properties is important, dis- course interpretation is needed as well. It is by un- derstanding the passage as a discourse that we see edgy, like a box, computery, and many buttons as descriptions of the type of design D does not pre- fer, and hand-held, organic shape, and simple de- signs as descriptions of the type he does. These de- scriptions are not in general synonyms/antonyms of one another; for example, there are hand-held “computery” devices and simple designs that are edgy. The unison/opposition among the descrip- tions is due to how they are used in the discourse. This paper focuses on such relations between the targets of opinions in discourse. Specifically, in this work, we propose a scheme of opinion frames, which consist of two opinions that are related by virtue of having united or opposed targets. We argue that recognizing opinion frames will pro- vide more opinion information for NLP applica- tions than recognizing individual opinions alone. Further, if there is uncertainty about any one of the components, we believe opinion frames are an ef- fective representation incorporating discourse in- formation to …

Content truncated to 500 words