Quantification of Biodiversity from Historical Survey Text with LLM-based Best-Worst-Scaling

Abstract

In this paper, we evaluate methods to determine biodiversity via quantity estimation from historical survey text. To that end, we formulate classification tasks and finally show that this problem can be successfully framed as regression based on best-worst-scaling with LLMs. We find that this approach is more cost effective and similarly robust to a fine-grained multi-class approach, allowing automated quantity estimation across species.