Experiments on Summary-based Opinion Classification

Proceedings of the NAACL HLT 2010 Workshop on Computational Approaches to Analysis and Generation of Emotion in Text, pages 107–115, Los Angeles, California, June 2010. c⃝2010 Association for Computational Linguistics Experiments on Summary-based Opinion Classification Elena Lloret Department of Software and Computing Systems University of Alicante Apdo. de Correos 99 E-03080, Alicante, Spain elloret@dlsi.ua.es Horacio Saggion Department of Infomation and Communication Technologies Grupo TALN Universitat Pompeu Fabra C/T´anger, 122-134, 2nd floor 08018 Barcelona, Spain horacio.saggion@upf.edu Manuel Palomar Department of Software and Computing Systems University of Alicante Apdo. de Correos 99 E-03080, Alicante, Spain mpalomar@dlsi.ua.es Abstract We investigate the effect of text summarisa- tion in the problem of rating-inference – the task of associating a fine-grained numerical rating to an opinionated document. We set-up a comparison framework to study the effect of different summarisation algorithms of various compression rates in this task and compare the classification accuracy of summaries and doc- uments for associating documents to classes. We make use of SVM algorithms to associate numerical ratings to opinionated documents. The algorithms are informed by linguistic and sentiment-based features computed from full documents and summaries. Preliminary re- sults show that some types of summaries could be as effective or better as full documents in this problem. 1 Introduction Public opinion has a great impact on company and government decision making. In particular, compa- nies have to constantly monitor public perception of their products, services, and key company represen- tatives to ensure that good reputation is maintained. Recent cases of public figures making headlines for the wrong reasons have shown how companies take into account public opinion to distance themselves from figures which can damage their public image. The Web has become an important source for find- ing information, in the field of business intelligence, business analysts are turning their eyes to the Web in order to monitor public perception on products, services, policies, and managers. The field of senti- ment analysis has recently emerged (Pang and Lee, 2008) as an important area of research in Natural Language Processing (NLP) which can provide vi- able solutions for monitoring public perception on a number of issues; with evaluation programs such as the Text REtrieval Conference track on blog min- ing 1, the Text Analysis Conference 2 track on opin- ion summarisation, and the DEfi Fouille de Textes program (Grouin et al., 2009) advances in the state of the art have been produced. Although sentiment analysis involves various different problems such as identifying subjective sentences or identifying posi- tive and negative opinions in text, here we concen- trate on the opinion classification task; and more specifically on rating-inference, the task of identify- ing the author’s evaluation of an entity with respect to an ordinal-scale based on the author’s textual eval- uation of the entity (Pang and Lee, 2005). The spe- cific problem we study in this paper is that of as- sociating a fine-grained rating (1=worst,...5=best) to a review. This is in general considered a dif- ficult problem because of the fuzziness inherent of …

Content truncated to 500 words