Large Language Models for Automatic Milestone Detection in Group Discussions

TL;DR

This paper explores the use of large language models, specifically GPT, for automatic detection of milestones in group oral discussions, addressing challenges like truncated utterances and variable response quality.

Contribution

It introduces a new group task experiment and demonstrates that iterative prompting of GPT with transcript chunks outperforms traditional semantic search methods.

Findings

Iterative GPT prompting improves milestone detection accuracy.

Processing transcripts in chunks outperforms embedding-based search.

GPT response quality varies with context window size.

Abstract

Large language models like GPT have proven widely successful on natural language understanding tasks based on written text documents. In this paper, we investigate an LLM's performance on recordings of a group oral communication task in which utterances are often truncated or not well-formed. We propose a new group task experiment involving a puzzle with several milestones that can be achieved in any order. We investigate methods for processing transcripts to detect if, when, and by whom a milestone has been completed. We demonstrate that iteratively prompting GPT with transcription chunks outperforms semantic similarity search methods using text embeddings, and further discuss the quality and randomness of GPT responses under different context window sizes.