With the rapid advancement of generative artificial intelligence technologies, virtual avatars have become increasingly personalized in appearance generation, behavioral mapping, and functional expansion. Rather than serving merely as visual projectio...
With the rapid advancement of generative artificial intelligence technologies, virtual avatars have become increasingly personalized in appearance generation, behavioral mapping, and functional expansion. Rather than serving merely as visual projections, AI-driven avatars have evolved into core interfaces through which users realize self-identity and emotional expression in digital spaces. However, the heightened level of personalization also introduces new psychological challenges, such as the gap between real and ideal selves, expressive burden in real-time interaction, and emotional depletion within social participation.
This study focuses on Generation Z users of virtual avatars, excluding AI-personified agents and asset-based generative systems, and examines how AI personalization influences users’ virtual identity (VI), thereby affecting self-expression (SE)and emotional resonance (ER), which in turn lead to creative behavior (CB)and behavioral intention for continuous use (BI). It further verifies the boundary-modulating role of emotional fatigue (EF)in this process.
Grounded in the Proteus Effect, Self-Congruence Theory, and Social Identity Theory (SIT), the study conceptualizes self-expression (SE) and emotional resonance (ER)as dual mediating mechanisms that translate identity into behavior, while defining emotional fatigue (EF)as a restrictive boundary mechanism within that translation. Accordingly, an integrated structural model was established—AI Personalization (AIP-V/B/F) → VI → SE/ER → CB/BI—with EF hypothesized to negatively moderate the VI → BI and SE → CB paths. Through case analysis, three representative platform mechanisms—Ready Player Me (static), ZEPETO (hybrid), and VTube Studio (dynamic)—were abstracted into appearance-driven, social-ecological, and dynamically driven modes of AI personalization, revealing variations in identity-to-behavior pathways across interaction structures.
The research proceeded in three stages. Pre-Study 1 (N = 307)developed the AI personalization (AIP-V/B/F) scale through literature synthesis, theoretical review, case analysis, and semi-structured interviews, followed by expert evaluation to ensure semantic and cultural validity, and exploratory factor analysis (EFA) confirming the three-dimension structure. Pre-Study 2 (N = 325)constructed the scales for psychological and behavioral outcome variables—VI, SE, ER, EF, CB, and BI—via EFA to refine items and ensure internal consistency. Empirical Study (N = 472)conducted confirmatory factor analysis (CFA) and structural equation modeling (SEM) to test model fit and path relationships. The moderating effects of EF were verified through interaction-term analysis and simple-slope tests within the SEM framework. Multi-group analysis (MGA) compared path strengths across static, hybrid, and dynamic platforms, controlling for gender, age, usage frequency, and experience.
Results showed that all three dimensions of AI personalization significantly enhanced virtual identity. VI significantly increased both SE and ER, which in turn positively influenced CB and BI, forming dual pathways of “identity → expression/resonance → behavior/intention.” EF negatively moderated the VI → BI and SE → CB paths, indicating that when emotional resources were depleted, the translation from identity to behavioral outcomes weakened. Multi-group results revealed platform-specific differences. in static platforms, AIP-V → VI was strongest, with appearance similarity and cross-platform consistency stabilizing identity, in hybrid platforms, social feedback and functional expansion sustained identity, and in dynamic platforms, AIP-B and AIP-F exerted the greatest effects, as real-time mapping and multimodal interaction intensified emotional resonance yet also elevated fatigue.
Theoretically, this study integrates AI personalization, virtual identity, the SE–ER dual mechanism, and emotional-load regulation into a unified model, addressing fragmentation and contextual limitations in prior research. Practically, it provides design implications by platform type. static platforms should strengthen AI-based appearance generation and cross-platform coherence to stabilize identity, hybrid platforms should leverage social feedback analytics and functional plug-in expansion to enhance sustained participation, and dynamic platforms should optimize motion prediction, lip-sync smoothing, and rhythm control to mitigate emotional fatigue in high-intensity interaction contexts.
Overall, through two rounds of scale development and one structural-equation validation, this study elucidates the mechanism of identity formation and behavioral translation in AI-personalized virtual avatars and provides both theoretical and practical foundations for sustainable AI-driven platform design.