Can LLMs benefit from research evaluation training data?

Investigate whether fine-tuning large language models with domain-specific research evaluation training data (e.g., article–score pairs or review judgements) improves their accuracy for article quality assessment, and quantify any gains across fields.

Background

While LLMs are broadly pre-trained on massive corpora, it is unknown if targeted fine-tuning with evaluation-specific data enhances their performance for research-quality tasks.

Answering this would guide whether collecting and curating such datasets is worthwhile and how they should be structured.