Redefining Toxicity: An Objective and Context-Aware Approach for Stress-Level-Based Detection

2025-03-20

AI Summary

Toxicity detection faces a fundamental challenge: the absence of an objective, quantifiable definition of “toxicity,” leading to low-quality subjectively annotated data and consequently poor model robustness and generalization. This paper proposes a novel toxicity definition framework grounded in physiologically measurable stress levels—specifically, multimodal physiological stress signals (e.g., heart rate variability, electrodermal activity)—as objective proxies for toxicity assessment. Building on this foundation, we develop a context-aware semantic representation learning method, construct a self-collected stress-annotated dataset, and design a tailored training paradigm. Experiments demonstrate that our approach significantly improves detection consistency (+23.6%) and cross-domain generalization on the proprietary dataset, effectively mitigating biases inherent in subjective annotation. By anchoring toxicity identification in reproducible, empirically verifiable physiological responses, this work establishes a rigorous, biologically grounded foundation for toxicity detection.

Technology Category

• Natural Language Processing: Safety and Robustness
• Computer Vision: Adversarial Attacks & Robustness
• Machine Learning: Adversarial Learning & Robustness

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Security and Privacy: Large-scale security measurements
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems

Abstract

The fundamental problem of toxicity detection lies in the fact that the term"toxicity"is ill-defined. Such uncertainty causes researchers to rely on subjective and vague data during model training, which leads to non-robust and inaccurate results, following the 'garbage in - garbage out' paradigm. This study introduces a novel, objective, and context-aware framework for toxicity detection, leveraging stress levels as a key determinant of toxicity. We propose new definition, metric and training approach as a parts of our framework and demonstrate it's effectiveness using a dataset we collected.

Problem

• Ill-defined toxicity term causes subjective model training.
• Non-robust results due to vague data in toxicity detection.
• Proposes stress-level-based framework for objective toxicity detection.

Innovation

• Context-aware framework for toxicity detection
• Stress levels as key toxicity determinant
• New definition, metric, and training approach