Recent advances in Large Language Models (LLMs) have transformed how we interact with and search for information online. At the Geographic Data Service we are investigating how these powerful tools might revolutionise research data discovery through our ESRC-funded ‘Talk data to me!’ project.
The Promise of LLMs in Research
LLMs offer exciting potential for helping researchers navigate the vast landscape of available datasets. By understanding natural language queries and context, these models could make finding relevant research data more intuitive and efficient than ever before. Our primary goal is to develop a semantic search tool that will enable researchers to seamlessly search across diverse UKRI research data infrastructure.
Understanding Researcher Perspectives
While the technical capabilities of LLMs are impressive, their adoption in research workflows raises important questions about acceptability and trust. To address these concerns head-on, we’ve been conducting comprehensive focus groups with key stakeholders in the research community:
- PhD Students: Early-career researchers who are digital natives but may have specific concerns about reliability
- Academic Researchers: Experienced scholars who need to maintain rigorous research standards
- Data Services Teams: Professionals who understand both technical requirements and user needs
- Non-Academic Users: Including local government analysts, third sector organizations, and institutions like the Office for National Statistics (ONS)
Looking Ahead
Our focus group discussions have yielded valuable insights into how different research communities view LLM technology and its potential role in data discovery. We’re excited to share detailed findings from this investigation in the new year, which will help shape the development of more effective and trustworthy research tools.
The results will inform our approach to building LLM-powered search tools that not only meet technical requirements but also address the real concerns and needs of the research community.
Stay tuned for our comprehensive findings report in early 2025.
