Predicting Team Performance from Communications in Simulated Search-and-Rescue

TL;DR

The study addresses predicting team performance from conversational data in a simulated search-and-rescue setting, where individual traits are not directly observable. It combines BEARD pre-trial profiles and TED process measures with unsupervised text analysis (Latent Dirichlet Allocation) of transcripts, followed by gap-statistics-based clustering to uncover interaction patterns and an early prediction/intervention pipeline. Key findings show BEARD/TED relationships to performance and demonstrate that an 8-cluster structure and a 12-topic model capture meaningful variance, with predictions reaching up to $76\%$ accuracy by about one-third into trials, enabling timely interventions. This work supports AI-enabled agents that monitor team conversations and intervene to improve outcomes in high-stakes, collaborative tasks.

Abstract

Understanding how individual traits influence team performance is valuable, but these traits are not always directly observable. Prior research has inferred traits like trust from behavioral data. We analyze conversational data to identify team traits and their correlation with teaming outcomes. Using transcripts from a Minecraft-based search-and-rescue experiment, we apply topic modeling and clustering to uncover key interaction patterns. Our findings show that variations in teaming outcomes can be explained through these inferences, with different levels of predictive power derived from individual traits and team dynamics.