Hate Detection API

Detect anti-semitic content and hate speech in short texts. Obfuscation-aware, bilingual (English + Hebrew), with single and batch analysis.

What does Hate Detection API do?

Hate Detection API classifies short texts (up to 280 characters — the length of a social media post) for anti-semitic content and general hate speech. It returns a structured decision (allow, flag, or block) along with confidence scores for both categories.

The classifier is obfuscation-aware, meaning it recognizes common evasion techniques such as leet speak (h4te), homoglyphs (visually similar Unicode characters), and spaced letters (h a t e). It supports English and Hebrew natively, with additional coverage for Arabic, Russian, and German.

Designed for content moderation pipelines, community platforms, and social media monitoring tools.

Actions

Action	Description
analyze	Analyze a single text for hate speech and anti-semitic content
analyze_batch	Analyze up to 100 texts in one request

Input

analyze — single text

{
  "action": "analyze",
  "text": "Text content to analyze here",
  "threshold": 0.8
}

Field	Type	Default	Description
text	string	required	Text to analyze (1–280 characters)
threshold	float (0.0–1.0)	0.8	Score above which the text is blocked. Lower values = stricter moderation.

analyze_batch — batch analysis

{
  "action": "analyze_batch",
  "texts": [
    { "id": "post-001", "text": "First post to analyze" },
    { "id": "post-002", "text": "Second post to analyze" }
  ],
  "threshold": 0.8
}

Field	Type	Default	Description
texts	object[]	required	List of items to analyze (1–100 items)
texts[].id	string	required	Unique identifier for this item
texts[].text	string	required	Text to analyze (1–280 characters)
threshold	float (0.0–1.0)	0.8	Blocking threshold applied to all items

Output

Single analyze

{
  "decision": "block",
  "scores": {
    "anti_semitic": 0.94,
    "hate_speech": 0.87
  },
  "confidence_label": "critical"
}

Batch analyze

{
  "results": [
    {
      "id": "post-001",
      "decision": "allow",
      "scores": { "anti_semitic": 0.02, "hate_speech": 0.05 },
      "confidence_label": "low"
    },
    {
      "id": "post-002",
      "decision": "flag",
      "scores": { "anti_semitic": 0.61, "hate_speech": 0.45 },
      "confidence_label": "medium"
    }
  ]
}

Decision values

Decision	Meaning
allow	Text is clean; no significant signals detected
flag	Borderline content; recommend human review
block	Score exceeds threshold; recommend blocking or removal

Confidence labels

Label	Score range
low	0.0 – 0.3
medium	0.3 – 0.6
high	0.6 – 0.85
critical	0.85 – 1.0

Features

• Three-tier moderation decisions — allow / flag / block for easy pipeline integration
• Anti-semitism and hate speech scores — two independent scores per text
• Obfuscation detection — recognizes leet speak, homoglyphs, and spaced characters
• Bilingual — English and Hebrew, with Arabic, Russian, and German support
• Adjustable threshold — tune strictness to your moderation policy
• Batch mode — analyze up to 100 texts per request
• No GPU required — heuristic classifier with fast cold starts

Limitations

• Maximum text length: 280 characters per item
• Maximum batch size: 100 texts per request
• The classifier uses heuristics and lexicon-based methods; it is not a fine-tuned LLM and may produce false positives or false negatives on highly nuanced content
• Best suited for social media post-length content; longer documents should be split into chunks

Usage on Apify

This actor is billed per Apify platform compute units. Analysis is fast — batch runs of 100 texts typically complete in under 10 seconds.