What Predicts Interpersonal Affect? Preliminary Analyses from Retrospective Evaluations

While the field of affective computing has contributed to greatly improving the seamlessness of human-robot interactions, the focus has primarily been on the emotional processing of the self, rather than the perception of the other. To address this gap, in a user study with 30 participant dyads, we collected the users' retrospective ratings of the interpersonal perception of the other interactant, after a short interaction. We made use of CORAE, a novel web-based open-source tool for COntinuous Retrospective Affect Evaluation. In this work, we analyze how these interpersonal ratings correlate with different aspects of the interaction, namely personality traits, participation balance, and sentiment analysis. Notably, we discovered that conversational imbalance has a significant effect on the retrospective ratings, among other findings. By employing these analyses and methodologies, we lay the groundwork for enhanced human-robot interactions, wherein affect is understood as a highly dynamic and context-dependent outcome of interaction history.