ESRA 2027 sessions by theme
Back to Overview of Sessions
LLMs for Open-Ended Text Coding: Progress, Opportunities and Ongoing Barriers |
| Coordinator 1 | Dr Liam Wright (University College London) |
| Coordinator 2 | Mr Matt Brown (University College London) |
| Coordinator 3 | Ms Erica Wong (University College London) |
| Coordinator 4 | Dr Laone Maphane (University of Southampton) |
Session Details
Open-ended text is necessary to collect certain information accurately from survey respondents, but coding such text at scale has historically been costly and slow, sometimes prohibitively. The advent of performant large language models (LLMs) changes this calculus. In so doing, it also lowers the barrier to bringing other non-standard text into survey research, which could enhance the utility of surveys further still: e.g., social media posts, essays, teacher reports, interviewer notes and observations.
This session invites contributions on any use of LLMs for coding or classifying open text in a survey context. We are interested in applications, evaluations (e.g., how accurate are LLMs and at what cost), and work tackling ongoing obstacles (e.g., using LLMs while meeting data protection laws), with a particular interest in proposals discussing the use of LLMs with non-standard text data sources.
As this is a fast-developing area, this session will emphasise sharing learning over finished results. Works in progress, negative findings, and reports of what did not work are welcome. The session will also include time for audience discussion of common barriers, practical solutions participants have found, and opportunities that remain underexplored.
