Computationally modeling interpersonal trust.

摘要: We present a computational model capable of predicting—above human accuracy—the degree trust person has toward their novel partner by observing the trust-related nonverbal cues expressed in social interaction. summarize our prior work, which we identify that signal untrustworthy behavior and also demonstrate mind’s readiness to interpret those assess trustworthiness robot. domain knowledge gained from work using human-subjects experiments, when incorporated into feature engineering process, permits outperform both predictions baseline built naivete' this knowledge. then construction hidden Markov models incorporate temporal relationships among cues. By interpreting resulting learned structure, observe emulate different levels exhibit sequences From observation, derived sequence-based features further improve accuracy model. Our multi-step research process presented paper combines strength experimental manipulation machine learning not only design but understanding dynamics interpersonal trust.