This post is from project team member Maarten Marx:
The book From Text to Political Positions. Text analysis across disciplines edited by Bertie Kaal, Isa Maks and Annemarie van Elfrinkhof recently appeared.
It contains chapters by DiLiPaD researchers Graeme Hirst and Maarten Marx.
From Text to Political Positions addresses cross-disciplinary innovation in political text analysis for party positioning. Drawing on political science, computational methods and discourse analysis, it presents a diverse collection of analytical models including pure quantitative and qualitative approaches. By bringing together the prevailing text-analysis methods from each discipline the volume aims to alert researchers to new and exciting possibilities of text analyses across their own disciplinary boundary.
The volume builds on the fact that each of the disciplines has a common interest in extracting information from political texts. The focus on political texts thus facilitates interdisciplinary cross-overs. The volume also includes chapters combining methods as examples of cross-disciplinary endeavours. These chapters present an open discussion of the constraints and (dis)advantages of either quantitative or qualitative methods when evaluating the possibilities of combining analytic tools.
Chapter: Sentiment Analysis in Parliamentary Proceedings
Authors: Steven Grijzenhout, Maarten Marx, Valentin Jijkoun.
Abstract of Chapter
This chapter addresses the question whether opinion-mining techniques can successfully be used to automatically retrieve political viewpoints from parliamentary proceedings. Two specific preprocessing tasks were identified and systematically evaluated: automatically determining subjectivity in the publications and automatically determining the semantic orientation of the subjective parts. A corpus of recent parliamentary proceedings was collected and a gold standard annotation was created on both subjectivity and orientation. Following this, a number of models based on subjectivity lexicons and machine-learning algorithms were evaluated. Machine-learning algorithms perform best, but methods based on subjectivity lexicons also provide promising results. Based on these results we can conclude that opinion-mining techniques applied to political data score just as well as the state of the art in other more traditional domains of opinion mining like product reviews and blogs.
Chapter: Text to Ideology or Text to Party Status?
Authors: Graeme Hirst; Yaroslav Riabinin; Jory Graham; Magali Boizot-Roche; Colin Morris.
Abstract of Chapter
Recent papers have used support-vector machines with word features to classify political texts by ideology. Our own work on this topic led us to hypothesize that such classifiers are sensitive not to expressions of ideology but rather to expressions of attack and defense, opposition and government. We test this hypothesis by training on one set of parliamentary speeches and testing on another in which party roles have been interchanged, and we find that the performance of the classifier completely disintegrates. Moreover, some features that are indicative of each party ‘swap sides’ with the change of government. Our results suggest that the language of attack and defense, of government and opposition, will dominate and confound any sensitivity to ideology in these kinds of classifiers.