Automatic Annotation of Dream Report’s Emotional Content with Large Language Models
Workshop on Computational Linguistics and Clinical Psychology January 1, 2024 Lorenzo Bertolini, V. Elce, Adriana Michalak et al.
Dream reports are typically analyzed by trained human annotators, a time-consuming process. While earlier natural language processing tools could automate some analysis, they could not reason over full report context, required extensive preprocessing, and were rarely validated against manual scoring. This study used large language models (LLMs) to replicate manual annotation of dream reports, focusing on references to emotions. An off-the-shelf LLM method performed poorly, likely due to linguistic differences between reports from different individuals. A bespoke text classification method achieved high performance and was robust against biases. The approach may enable analysis of large dream datasets and improve reproducibility across studies.