Abstract: [Purpose/Significance] As a social bot, virtual companions have shown significant potential in delivering mental health support to student populations. This study aims to reveal impact of the humor style employed by virtual companions on the alleviation of academic stress and to identify its underlying mechanisms, thereby providing empirical evidence for the development of anthropomorphic language design in virtual companion interactions. [Design/Methodology] An inter-group controlled experiment was conducted based on a one-factor between-subjects design, manipulating the virtual companions' humor style (self-enhancing vs. affiliative). University students were recruited to interact with a virtual companion system designed to alleviate stress, while their perceived social support, sense of intimacy, stress level changes, and willingness to continue using the virtual companion system were measured by validated scales. [Findings/Conclusion] Compared with the self-enhancing type, the affiliative type of humor style is a more effective anthropomorphism strategy for artificial intelligence language design. Virtual companions with an affiliative type of humor style can bring about greater perceived stress reduction by enhancing users' perceived social support, whereas increasing the intimacy and their willingness to use the virtual companion. Moreover, the individual field cognitive style plays a significant moderating role, that is, field-dependent users will experience higher perceived social support when interacting with virtual companions using affiliative type of humor style. [Originality/Value] It enriches the research on virtual companions within the domain of mental health, providing both theoretical insights and practical guidance for the anthropomorphic language design in AI systems.

Keywords: Virtual companion, Anthropomorphism, Humor style, Academic stress, Field cognitive style