EgoSocial: Benchmarking Proactive Intervention Ability of Omnimodal LLMs via Egocentric Social Interaction Perception
This work addresses the need for AI assistants in AR/VR to better understand social dynamics for timely interventions, though it is incremental as it builds on existing omnimodal LLMs with a new dataset and method.
The paper tackles the problem of AI lacking social awareness for proactive intervention in egocentric social interactions by introducing the EgoSocial dataset and EgoSoD method, resulting in improvements such as boosting Phi-4 by 45.6% and Gemini 2.5 Pro by 9.9% on intervention timing performance.
As AR/VR technologies become integral to daily life, there's a growing need for AI that understands human social dynamics from an egocentric perspective. However, current LLMs often lack the social awareness to discern when to intervene as AI assistant. This leads to constant, socially unaware responses that may disrupt natural conversation and negatively impact user focus. To address these limitations, we introduce EgoSocial, a large-scale egocentric dataset with 13,500 social video-question pairs, specifically designed to benchmark intervention in social interaction perception. We also present an in-depth analysis of current omnimodal LLMs (OLLMs) to assess their effectiveness in detecting diverse social contextual cues. Experiments show that OLLMs still struggle to detect the intervention timing (14.4% for Gemini 2.5 Pro). We also propose EgoSoD (EgoSocial Detection), an end-to-end method for robustly discerning social dynamics. Informed by our OLLM analysis, EgoSoD integrates multimodal contextual cues (e.g., audio and visual cues) into a social thinking graph, dynamically modeling participants and interactions. Our method proactively detects intervention timing and social interactions, precisely determining when to intervene. Our EgoSoD improves Phi-4 by 45.6% and Gemini 2.5 Pro by 9.9% on Intervention Timing performance, and improves Phi-4 by 20.4% and Gemini 2.5 Pro by 6.9% on overall Social Interaction performance. We will release the dataset and code soon.