KJCKorea
Journal Central

This study examines how proactive intervention methods by conversational AI agents affect user experience in dyadic home conversations. AI Agent interventions in social contexts create a "proactivity dilemma" by influencing conversational dynamics and relational norms. A 3 × 2 within-subject experiment (N=30) measured social appropriateness, agency, and pleasantness across conversation types and intervention methods. Consent-based interventions significantly outperformed direct interventions, particularly for agency in opinion conflicts. Mediation analysis validated the pathway: intervention method → social appropriateness → agency → pleasantness. Qualitatively, consent-based interventions represented "knocking" versus direct interventions as "intrusion." This research establishes a proactive AI Agent acceptance model for social contexts with consent-based design as default.

This study examined how an interactive conversational AI agent’s emotion-regulation intervention shapes pre-judgment appraisals and approach motivation in high-anxiety decisions. Grounded in cognitive appraisal dimensions (certainty, perceived control, anticipated effort) and the Appraisal Tendency Framework (ATF), we presented a webtoon-based scenario and ran a 2×2 experiment (strategy: situation selection vs. attentional deployment; style: rational vs. empathic). Participants rated outcomes on 7-point scales. Two-way ANOVA showed a significant strategy × style interaction for approach motivation: under situation selection, rational expression increased approach motivation. Certainty showed significant main effects of strategy and style. Perceived control was not significant but showed a directional interaction pattern, and anticipated effort showed no differences. Overall, emotion regulation appears to alter appraisal conditions before judgment rather than only easing emotions after the fact. The findings suggest conversational AI can act as an external mediator in this appraisal process and that emotion-regulation mechanisms can be implemented and tested through technology-mediated interaction.

With the advancement of large language models(LLMs), in-context learning has become a key approach, driving research on prompting techniques. Among them, few-shot Chain-of-Thought(CoT) prompting, which induces explicit reasoning, shows strong performance but depends on example composition. While prior work focused on diversity or uncertainty in example selection, difficulty-based approaches remain underexplored. This study proposes a method to identify high-difficulty questions by combining pairwise difficulty comparisons conducted by an LLM with a Swiss tournament structure, constructing few-shot CoT exemplars with human reasoning annotations. Experiments on 1,319 GSM8K problems show that the proposed method outperforms random, uncertainty-based, and direct difficulty evaluation approaches by 2.12%p, 1.36%p, and 10.16%p, respectively.