EnsembleToxicityDetector<T>

Models & Types · AiDotNet.Safety.Text

Combines multiple toxicity detectors into a weighted ensemble for improved accuracy.

Inheritance: ModelBase<T, Vector<T>, Vector<T>> → SafetyModuleBase<T> → TextSafetyModuleBase<T> → EnsembleToxicityDetector<T>

For Beginners

Just like a panel of judges gives better verdicts than a single judge, combining multiple toxicity detectors gives more accurate results. This module runs several different detection approaches and combines their opinions.

How It Works

Aggregates findings from multiple toxicity detection strategies (rule-based, embedding-based, classifier-based) using configurable weights. The ensemble approach reduces both false positives and false negatives compared to any single detector.

References:

  • Ensemble methods for robust hate speech detection (ACL 2024)
  • MetaTox knowledge graph for enhanced LLM toxicity detection (2024, arxiv:2412.15268)
  • GPT-4o/LLaMA-3 zero-shot and few-shot hate speech detection (2025, arxiv:2506.12744)

Constructors

Constructor Summary
EnsembleToxicityDetector(ITextSafetyModule<T>[] detectors, double[] weights, double threshold) Initializes a new ensemble toxicity detector with custom sub-detectors and weights.
EnsembleToxicityDetector(double threshold) Initializes a new ensemble toxicity detector with default sub-detectors.

Properties

Property Summary
string ModuleName { get; }

Methods

Method Summary
IReadOnlyList<SafetyFinding> EvaluateText(string text)