Computationally modeling interpersonal trust.

Area of Science:

• Computational Social Science
• Human-Computer Interaction
• Machine Learning

Background:

• Human trust assessment relies on interpreting nonverbal cues.
• Prior research identified nonverbal indicators of untrustworthy behavior.
• The human mind readily interprets these cues, even in human-robot interactions.

Purpose of the Study:

• To develop a computational model for predicting interpersonal trust levels.
• To enhance trust prediction accuracy by integrating domain knowledge and temporal dynamics.
• To deepen the understanding of trust dynamics in social interactions.

Main Methods:

• Utilized domain knowledge from human-subject experiments for feature engineering.
• Constructed hidden Markov models to analyze temporal relationships in nonverbal cues.
• Developed sequence-based temporal features to improve model performance.

Main Results:

• The computational model achieved above-human accuracy in predicting trust.
• Incorporating domain knowledge significantly outperformed baseline and human predictions.
• Learned temporal structures from hidden Markov models revealed distinct trust-related nonverbal cue sequences.

Conclusions:

• Computational models can accurately predict interpersonal trust by analyzing nonverbal cues.
• Integrating experimental domain knowledge and temporal analysis is crucial for robust trust prediction.
• This research advances both computational trust modeling and the understanding of interpersonal trust dynamics.