Title: Outsmarting the Chameleon: Counterfactual Decoupling for Tactical OOD Shifts in Live Streaming Risk Assessment

Original: arXiv:2606.02946v1 Announce Type: new Abstract: Live streaming has emerged as a primary medium for social interaction and digital commerce, yet it is increasingly plagued by sophisticated risks. A fundamental challenge in this domain is \emph{tactical out-of-distribution (OOD) shift}: while malicious actors maintain stable underlying objectives, they continuously redesign narrative packaging to evade detection. Such adversarial shifts expose critical limitations of existing OOD generalization paradigms, whose assumptions are difficult to satisfy in the presence of tightly coupled intent-tactic evolution and ill-defined raw-level counterfactuals. In this paper, we tackle this issue from a \emph{latent causal} perspective and propose \underline{L}atent-\underline{P}redictive \underline{C}ounterfactual \underline{D}ecoupling~(LPCD), a plug-in framework for robust live streaming risk assessment. LPCD enables counterfactual reasoning under adversarial tactical re-packaging by modeling intent and narrative variation at the latent level, and enforces \emph{latent counterfactual consistency} to anchor risk prediction on causally stable malicious intent. At inference time, LPCD applies a lightweight, parameter-free calibration to further mitigate tactic-induced distribution shifts. Extensive experiments on large-scale industrial datasets and online production traffic demonstrate that LPCD consistently outperforms state-of-the-art baselines, validating its effectiveness in moderating evolving adversarial risks in real-world live streaming. The project page is available at https://qiaoyran.github.io/LiveStreamingRiskAssessment/.

Rewrite: Live streaming has become a dominant platform for both social engagement and e-commerce, but it is increasingly beset by complex and advanced threats. A significant hurdle in this field is the \emph{tactical out-of-distribution (OOD) shift}: although bad actors keep their core goals constant, they constantly alter their storytelling methods to avoid being caught. These adversarial changes highlight the shortcomings of current OOD generalization models, which struggle to meet their own assumptions when intent and tactics evolve together and raw-level counterfactuals are poorly defined. To address this, we approach the problem through a \emph{latent causal} lens and introduce \underline{L}atent-\underline{P}redictive \underline{C}ounterfactual \underline{D}ecoupling~(LPCD), a modular framework designed to enhance the robustness of live streaming risk assessment. LPCD facilitates counterfactual reasoning even when adversaries repackaging their tactics by capturing intent and narrative variations at the latent level. It also enforces \emph{latent counterfactual consistency}, ensuring that risk predictions are grounded in causally stable malicious intent. During inference, LPCD utilizes a simple, parameter-free calibration step to further reduce distribution shifts caused by tactical changes. Our extensive tests on large-scale industrial data and live production traffic show that LPCD consistently beats state-of-the-art baselines, proving its ability to handle evolving adversarial risks in real-world live streaming environments. More details can be found at https://qiaoyran.github.io/LiveStreamingRiskAssessment/.

Source: arXiv Generated at: 2026-06-03 00:00:00 UTC