Automatic Estimation of Interpersonal Engagement During Naturalistic Conversation Using Dyadic Physiological...

Area of Science:

• Psychophysiology
• Human-Computer Interaction
• Machine Learning

Background:

• Dyadic physiological responses offer insights into interpersonal dynamics like engagement.
• Previous research primarily focused on group-level analysis of physiological data.
• Dynamic, individual-level estimation of dyadic engagement remains underexplored.

Purpose of the Study:

• To develop and validate a machine learning model for dynamically estimating interpersonal engagement in dyads.
• To investigate the utility of individual and synchrony-based physiological features for engagement estimation.
• To assess the model's performance against baseline estimators and human observers.

Main Methods:

• Collected physiological data (ECG, skin conductance, respiration, skin temperature) from 16 dyads during naturalistic conversations.
• Extracted individual and synchrony-based physiological features.
• Employed multiple regression algorithms with leave-interval-out and leave-dyad-out crossvalidation to estimate self-reported engagement.

Main Results:

• Machine learning models achieved moderate accuracy in estimating dyadic engagement.
• Leave-dyad-out crossvalidation showed slightly higher accuracy than a baseline and outperformed human observers.
• Synchrony features and personality traits influenced estimation accuracy, with their removal negatively impacting performance.

Conclusions:

• Physiological measurements combined with machine learning can dynamically estimate interpersonal engagement during conversations.
• This approach holds potential for applications in conversation monitoring and enhancement.
• Further research is needed to contextualize estimation errors and determine acceptable accuracy levels.