Thesis Title
Forecasting of Listener Reaction Trajectory in Dialogues

Research Description
Anticipating a listener’s emotional and behavioral reactions is essential for planning effective and empathetic speech. Existing studies overlook its unique properties, including strong dependence on individual listener traits, along with delayed and evolving responses. This project formulates the task of listener reaction trajectory forecasting, develops methods for personalized prediction, and explores its downstream applications such as improving emotional support conversations, predicting conversational derailments, and strengthening large language model safety. By enabling proactive anticipation of reactions, the proposed research aims to enhance the safety, effectiveness, and appropriateness of conversational systems.

Supervisors
Prof Chris Leckie, University of Melbourne