This project is developed within the Oxford Computational Political Science Group (OCPSG) and brings together researchers working on computational political science, political text analysis, benchmarking, and artificial intelligence.

The benchmark initiative is focused on the development of transparent, reproducible, and multilingual infrastructure for the evaluation of large language models and fine-tuned models in policy agenda annotation for parliamentary speeches.

Project context

The project sits within the wider mission of OCPSG to advance computational methods in political science and public policy through interdisciplinary collaboration, methodological innovation, and research infrastructure.

Core team

Bastián González-Bustamante is a Postdoctoral Researcher in Computational Social Science at Leiden University and an Associate Professor of Public Administration at Universidad Diego Portales. He holds a DPhil from the University of Oxford, and he works on building large-scale text-as-data pipelines, deploying AI and machine learning models, and applying causal inference strategies in downstream analyses. He has published in Social Science Computer Review, Public Opinion Quarterly, World Development, Artificial Intelligence and Law, Government and Opposition, and elsewhere. He is currently working on an NWO-funded project using LLMs to analyse sustainable finance flows. In parallel, he contributes to the COST CoREx project on executive-bureaucratic careers and leads an Enlace-UDP project on NLP and LLMs to examine cabinet politics in presidential democracies. https://github.com/bgonzalezbustamante.

Tom Bellens is a Postdoctoral Researcher in Computational Political Science at the University of Antwerp. He obtained his PhD at the Public Governance Institute of KU Leuven, where his research examined political advisers and their career backgrounds using a text-as-data approach. He also holds a Master’s degree in Artificial Intelligence from KU Leuven. He focuses on substantive questions about political and administrative elites. Since 2026, he has worked on the DEMO-LIES project at the University of Antwerp, where he applies NLP methods to detect and analyse allegations of lying in parliamentary debates across more than 20 countries. He is also involved in the COST CoREx project, a comparative research network covering 35 countries. https://github.com/Tombellens.

Christopher Klamm is an interdisciplinary researcher with a background in computer science and political science. Having studied in Darmstadt and Zurich, he is now pursuing a PhD at the University of Mannheim and is affiliated with the Cologne Center of Comparative Politics. His research focuses on analysing rhetoric and framing in political communication using LLMs as interview agents in survey research and issues related to data quality and validation in the machine learning age. As a community builder in the field of computational social science, Christopher co-organises the tada.cool speaker series and actively participates in open science initiatives such as BLOOM (BigScience/Hugging Face), Expedition Aya (Cohere4AI), and the Data Provenance Initiative. https://github.com/chkla.

Marta Koch is a PhD Researcher at Imperial College London and a UN research consultant specialising in data science and machine learning methods. She works at the science-policy interface as a UNFCCC Technology Executive Committee Activity Groups Member, Scientific Expert Reviewer for UNEP’s Global Environment Outlook, WMO/UNEP IPCC and UNEP/UNESCO/FAO/UNDP Intergovernmental Science-Policy Platform (IPBES) and UNFCCC, UNEP, UNDESA and ITU delegate. She has published in International Affairs, Science of the Total Environment, The Innovation and Nature Portfolio journals and UNEP Global Environment Outlook (GEO). Her work has been endorsed under UNESCO’s International Decade of Sciences for Sustainable Development. She has authored publications for UNOPS, UN-Habitat, UNDP, UN Climate Technology Centre and Network (CTCN), LSE, UCL, IISD, and UN SDSN. https://github.com/MartaKoch.