The browser you are using is not supported by this website. All versions of Internet Explorer are no longer supported, either by us or Microsoft (read more here: https://www.microsoft.com/en-us/microsoft-365/windows/end-of-ie-support).

Please use a modern browser to fully experience our website, such as the newest versions of Edge, Chrome, Firefox or Safari etc.

The matter of pre-training in political studies: More data better than domain-specific new study shows

Black and white photo of Annika Fredén. Photo.

In this study we elaborate the trade-off between pre-training models on domain-specific data versus a more general dataset in a political context. We start from motions from parliamentarians in the Swedish Riksdag, and focus on the two competitors for government office parties: Social Democrats and Moderates.

Elaborating word embeddings on central keywords related to issues and ideology based on voters’ perceptions of parties, we find that the word embeddings that started from pre-training on a bigger dataset provided clearer ideological differences between the parties. This article has been published in Journal of Elections, Public Opinion and Parties and is co-work between political science, Lund and computer science, Chalmers.

Why It Matters:
The study’s implications go beyond Sweden. It highlights how AI tools can assist political scientists and policymakers worldwide in dissecting complex political discourse, identifying ideological trends, and better understanding voter communication. This approach could offer new ways to track political polarization, predict policy shifts, and even help voters make more informed decisions.

 Authors: Annika Fredén, Moa Johansson, Denitsa Saynova

Title: Word embeddings on issues and ideology from Swedish parliamentarians’ motions: A Comparative Approach

Link: Full article: Word embeddings on ideology and issues from Swedish parliamentarians’ motions: a comparative approach