Abstract: Using a systematic literature review that combines quantitative evidence synthesis and qualitative theoretical reconstruction, this study reviews emotional recovery in AI service failure contexts. Based on the Stimulus-Organism-Response (S-O-R) framework, AI emotional recovery strategies are classified into emotional expression and empathic response. Their mechanisms, effects, and boundary conditions are examined from the perspectives of social cognition, social interaction, emotional and attitudinal outcomes, behavioral intentions, and user, situational, and technological characteristics. The findings show that these strategies can alleviate negative emotions and enhance forgiveness, satisfaction, and continuance intention, although their effectiveness varies across users and service contexts. This study expands the research perspectives on AI service recovery and offers implications for the human-centered design of intelligent service systems.

Key words: Artificial intelligence, Information service, AI service failure, AI service recovery, Human-AI interaction, Affective computing