HarassGuard: Detecting Harassment Behaviors in Social Virtual Reality with Vision-Language Models

2026-04-01

AI Summary

This work addresses the prevalent issue of physical harassment in social virtual reality, where existing safeguards often rely on post-hoc responses or sensitive biometric data, raising significant privacy concerns. To mitigate these risks, the authors propose a real-time detection method that operates solely on visual inputs, leveraging a vision-language model (VLM) enhanced through prompt engineering and fine-tuning. This approach enables context-aware harassment recognition without requiring any biometric information. Evaluated on an IRB-approved visual dataset, the method achieves performance comparable to state-of-the-art baselines using only 200 training samples, attaining 88.09% accuracy in binary classification and 68.85% in multi-class settings. The proposed framework thus offers a privacy-preserving solution that effectively balances contextual reasoning with computational efficiency.

Abstract

Social Virtual Reality (VR) platforms provide immersive social experiences but also expose users to serious risks of online harassment. Existing safety measures are largely reactive, while proactive solutions that detect harassment behavior during an incident often depend on sensitive biometric data, raising privacy concerns. In this paper, we present HarassGuard, a vision-language model (VLM) based system that detects physical harassment in social VR using only visual input. We construct an IRB-approved harassment vision dataset, apply prompt engineering, and fine-tune VLMs to detect harassment behavior by considering contextual information in social VR. Experimental results demonstrate that HarassGuard achieves competitive performance compared to state-of-the-art baselines (i.e., LSTM/CNN, Transformer), reaching an accuracy of up to 88.09% in binary classification and 68.85% in multi-class classification. Notably, HarassGuard matches these baselines while using significantly fewer fine-tuning samples (200 vs. 1,115), offering unique advantages in contextual reasoning and privacy-preserving detection.

Problem

• harassment detection
• social virtual reality
• privacy-preserving
• online safety
• vision-language models

Innovation

• vision-language model
• harassment detection
• social virtual reality
• privacy-preserving AI
• prompt engineering