BROWSE
Detail
Conference Paper
Social VR Administrator: A Context-Aware Model Agent Design for Harassment Prevention in Social VR
Cited 0 time in 
Share 
Share
- Authors
- Junhee Lee, Seungwon Woo, Hwanjo Heo, Jinwoo Kim
- Issue Date
- 2026-03
- Citation
- Conference on Virtual Reality and 3D User Interfaces Abstracts and Workshops (VRW) 2026, pp.39-44
- Publisher
- IEEE
- Language
- English
- Type
- Conference Paper
- Abstract
- As Social VR platforms continue to expand, ensuring user safety against interpersonal harassment has become an increasingly critical challenge. Existing mitigation approaches - largely dependent on human moderation or invasive biometric sensing - suffer from limited scalability, privacy concerns, and insufficient contextual understanding. In this paper, we present Social VR Administrator (SVRA), a context-aware intelligent agent framework designed to provide cost-effective, transparent, and automated moderation in social VR environments. SVRA employs a novel modular architecture comprising three key components: A Context Analyzer that interprets behavioral norms by incorporating environmental settings and social relationships; a Dynamic Model Selector that adaptively balances detection accuracy and computational efficiency by escalating analysis from lightweight screening models to advanced Vision Language Models (VLMs); and an Action Module that executes graduated interventions ranging from warnings to blocking. Experimental results demonstrate that SVRA effectively balances detection accuracy with resource consumption, enabling a robust, privacy-preserving harassment prevention.
- Keyword
- Agentic AI, Harassment Detection, Intelligent Agent, Social VR, VR Security
- KSP Keywords
- Agent design, Agent framework, Computational Efficiency, Context-aware model, Detection accuracy, Dynamic Models, Key Components, Model agent, Modular Architecture, Privacy concerns, Privacy-preserving