Political Analysis Advance Access originally published online on October 14, 2008
Political Analysis 2008 16(4):428-446; doi:10.1093/pan/mpn006
| ||||||||||||||||||||||||||||||||||||||||||||||||||||
This article appears in the following Political Analysis issue: Special Issue: The Statistical Analysis of Political Text [View the issue table of contents]
Parsing, Semantic Networks, and Political Authority Using Syntactic Analysis to Extract Semantic Relations from Dutch Newspaper Articles
Faculty of Social Sciences, Vrije Universiteit Amsterdam, De Boelelaan 1081, 1081 HV Amsterdam, the Netherlands
e-mail: wouter{at}vanatteveldt.com (corresponding author)
Amsterdam School of Communication Research, University of Amsterdam, Kloveniersburgwal 48, 1012 CX Amsterdam, the Netherlands
e-mail: p.c.ruigrok{at}uva.nl
Analysis of political communication is an important aspect of political research. Thematic content analysis has yielded considerable success both with manual and automatic coding, but Semantic Network Analysis has proven more difficult, both for humans and for the computer. This article presents a system for an automated Semantic Network Analysis of Dutch texts. The system automatically extracts relations between political actors based on the output of syntactic analysis of Dutch newspaper articles. Specifically, the system uses pattern matching to find source constructions and determine the semantic agent and patient of relations, and name matching and anaphora resolution to identify political actors. The performance of the system is judged by comparing the extracted relations to manual codings of the same material. Results on the level of measurement indicate acceptable performance. We also estimate performance at the levels of analysis by using a case study of media authority, resulting in good correlations between the theoretical variables derived from the automatic and manual analysis. Finally, we test a number of substantive hypotheses with regression models using the automatic and manual output, resulting in highly similar models in each case. This suggests that our method has sufficient performance to be used to answer relevant political questions in a valid way.
Author's note: Wouter van Atteveldt is an interdisciplinary Ph.D. student at the Departments of Communication Science and Artificial Intelligence at the Vrije Universiteit Amsterdam (the Netherlands). He hopes to defend his Ph.D. thesis on Methods for automatically extracting and representing media data in the Fall of 2008. Jan Kleinnijenhuis is a professor of mass communication. His research deals with news selection and news effects. Nel Ruigrok is a researcher at the Netherlands News Monitor and a research fellow in the Amsterdam School of Communications Research at the University of Amsterdam. Her research interests include the role of media in the political arena, especially during times of conflict.