Navigation auf


Department of Computational Linguistics Text Technologies

NCCR Democracy: How media report on political institutions beyond the state – a computer-based content analysis


We analyse newspapers and online media texts automatically. Until recently, most of the annotations in social science projects, such as classification of media content into analytical categories, were made by humans. Our current projects aim to implement a largely automated approach, based on computational linguistics, for the selection of media documents as well as for the classification of content. The project (1) uses well-established methods for tasks such as named-entity recognition and keyword detection (2) applies state-of-the-art text mining methods from biomedical to a new domain, (3) develops genuinely new techniques to current computational linguistics challenges, such as the integration of polarity detection in relation mining and automated approaches to framing.

Work in these projects is based on a list of governance organisations identified by the co-operating thematic projects and includes the following tasks:

1.   Selection of articles: using bibliographical databases we collect and pre-process the texts.

2.   Nomination and salience analysis: Named entity recognition and statistical analysis of frequencies are applied to obtain keyword and salience figures. These tasks are frequently used and well-understood in computational linguistics research.

3.   Tonality, sentiment detection and opinion mining: On the basis of our considerable experience in sentiment analysis, we extend sentence-level approaches to the relation level in order to find out who has a positive or negative attitude towards the entity of interest (e.g. governance organisations under scrutiny in the individual projects). To do so, relation mining approaches need to be included, for which we draw on our extensive previous work in the biomedical domain where we combine syntactic parsing and machine-learning approaches. Syntactic approaches are particularly suitable for the detection of polarity of relations.

4.   Frames: Numerous approaches in political communication have suggested to use framing devices for the analysis of political discourse. While these devices are suitable for manual analyses, only very few framing approaches have been tested in automatic procedures. Frame analysis is an experimental and novel research task for computational linguistics.


Further details about the nccr democracy can be found on the project homepageand especially on our sub-project website.

NCCR Democracy at the Department of Computational Linguistics at UZH

Project co-lead

  • Bruno Wüest (Institute for Political Science, University of Zurich)

The project is funded by the Swiss National Science Foundation (SNF) and runs 2013-2017.