The growing machine learning literature makes it difficult to identify strong baselines and determine which reported results can be meaningfully compared. Platforms such as Papers with Code have partially addressed this problem, but reliably extracting and organizing results from research papers remains challengingdue to the variability of experiments.

This thesis will involve extraction of results from tables as well as putting them into context based on the remainder of the paper by using large language models for the extraction process. The thesis’ output will be integrated into the existing Graph Learning Benchmarks platform and we will use graph machine learning benchmarks as the application domain.

Starting literature:

• A Position Paper on the Automatic Generation of Machine Learning Leaderboards (EMNLP 2025)
• Open Graph Benchmark (2020)
• Benchmarking Graph Neural Networks (JMLR 2023)