- Machine Learning-Based Content Analysis: Automating the analysis of frames and agendas in political communication research
- Award date
- 5 October 2016
- Number of pages
- Document type
- PhD thesis
- Faculty of Social and Behavioural Sciences (FMG)
- Amsterdam School of Communication Research (ASCoR)
We used machine learning to study policy issues and frames in political messages. With regard to frames, we investigated the automation of two content-analytical tasks: frame coding and frame identification. We found that both tasks can be successfully automated by means of machine learning techniques. Frame coding can be automated through supervised machine learning (SML). Results show that the performance of SML-based frame coding approaches the performance of human coders.
Furthermore, we have shown that frames can be automatically identified through clustering, a form of unsupervised machine learning. We used this method to identify issue frames in the nuclear power debate. We found that automatically identified frames closely resemble frames that have been identified in previous studies, by means of qualitative approaches.
In addition, we have shown that policy issues can be coded by means of SML as well as through semi-automatically created dictionaries. Again, automatic coding approaches the performance of human coders. Moreover, we demonstrated that SML and dictionary-based coding can be applied to different types of political messages (e.g., news articles and parliamentary records).
- Research conducted at: Universiteit van Amsterdam