Journal
DATA & KNOWLEDGE ENGINEERING
Volume 90, Issue -, Pages 38-53Publisher
ELSEVIER
DOI: 10.1016/j.datak.2013.07.003
Keywords
Topic models; Political science; Text Analysis
Ask authors/readers for more resources
Automatic content analysis is more and more becoming an accepted research method in social science. In political science researchers are using party manifestos and transcripts of political speeches to analyze the positions of different actors. Existing approaches are limited to a single dimension, in particular, they cannot distinguish between the positions with respect to a specific topic. In this paper, we propose a method for analyzing and comparing documents according to a set of predefined topics that is based on an extension of Latent Dirichlet Allocation (LDA) for inducing knowledge about relevant topics. We validate the method by showing that it can guess which member of a coalition was assigned a certain ministry based on a comparison of the parties' election manifestos with the coalition contract. We apply the method to German National Elections since 1990 and show that the use of our method consistently outperforms a baseline method that-simulates manual annotation of individual sentences based on keywords and standard text comparison. In our experiments, we compare two different extensions of LDA and investigate the influence of the used seed set. Finally, we give a brief illustration of how the output of our method can be interpreted to compare positions towards specific topics across several parties. (C) 2013 Elsevier B.V. All rights reserved.
Authors
I am an author on this paper
Click your name to claim this paper and add it to your profile.
Reviews
Recommended
No Data Available