This site hosts the benchmarking and documentation infrastructure for OCPSG Benchmarking LLMs, a project developed within the Oxford Computational Political Science Group (OCPSG).

OCPSG is a research initiative supported by Oxford’s Department of Politics and International Relations. It advances the use of computational methods, including machine learning and AI, in political science and policy through an interdisciplinary and collaborative research environment. Founded in 2024, the group has developed as an international research network committed to rigorous, evidence-based computational approaches to major political and policy challenges.

Why this project exists

This website focuses on one specific line of work within that broader mission: the development of transparent, reproducible, and multilingual benchmark infrastructure for policy agenda annotation in parliamentary speeches.

The project is designed to support systematic comparison across large language models, fine-tuned models, and related computational approaches used for political text classification. Its aim is not only to compare models, but also to improve transparency, comparability, and methodological rigour in the use of AI for political text analysis.

Why this fits OCPSG

This benchmark reflects OCPSG’s wider commitment to combining methodological innovation with substantive political research, training computational talent, and producing work that is both academically rigorous and useful beyond academia.

The group’s mission emphasises interdisciplinary collaboration, computational training, and the translation of technical research into accessible and decision-relevant outputs for wider audiences. In that spirit, this benchmark is intended as shared research infrastructure: a practical resource for scholars, students, and collaborators working on multilingual political texts, policy classification, and the evaluation of AI systems in social science.

Current focus

At present, the project is centred on:

• silver-standard benchmarking workflows;
• multilingual parliamentary speech data;
• standardised evaluation of LLMs and fine-tuned classifiers;
• reproducible submissions and documentation.

Contact

For general information about the wider Oxford Computational Political Science Group, see the OCPSG page at the Department of Politics and International Relations, University of Oxford.