Moderation Policies and Rules - Content Moderation

Moderation Policies and Rules

Moderation policies define what the AI engine checks for when analyzing content. A policy is a named collection of rules, each targeting a specific content concern.

Text Rules

Rule	Description
`profanity`	Detects profanity and vulgar language in transcripts
`hate_speech`	Identifies hate speech, slurs, and discriminatory language
`pii`	Flags personally identifiable information (names, emails, phone numbers, SSNs)
`violence_text`	Detects descriptions of violence or threats in spoken/written content
`sexual_text`	Identifies sexually explicit or suggestive language
`self_harm`	Flags references to self-harm or suicidal content
`dangerous_activities`	Detects instructions for dangerous or illegal activities
`custom_keywords`	Matches against a configurable list of blocked terms

Visual Rules

Rule	Description
`nudity`	Detects nudity or near-nudity in video frames
`violence_visual`	Identifies graphic violence, blood, or weapons in frames
`drugs`	Detects drug paraphernalia or substance use
`alcohol_tobacco`	Flags alcohol or tobacco product visibility
`offensive_symbols`	Identifies hate symbols, offensive gestures, or extremist imagery
`text_overlay`	Reads and moderates text overlaid on video frames
`child_safety`	Detects potential child exploitation or endangerment content

Predefined Policies

The platform includes two predefined policies:

Policy	Rules Included
`default_moderation_v2`	profanity, hate_speech, pii, nudity, violence_visual, violence_text, sexual_text
`strict_moderation_v1`	All text rules + all visual rules

Custom Policies

Create custom policies by selecting specific rules and configuring severity thresholds:

curl -X POST "$KALTURA_SERVICE_URL/service/moderationPolicy/action/add" \
  -d "ks=$KALTURA_KS" \
  -d "format=1" \
  -d "policy[name]=Corporate Content Policy" \
  -d "policy[rules][0][type]=profanity" \
  -d "policy[rules][0][severity]=high" \
  -d "policy[rules][1][type]=pii" \
  -d "policy[rules][1][severity]=medium" \
  -d "policy[rules][2][type]=custom_keywords" \
  -d "policy[rules][2][keywords]=competitor1,competitor2,internal-codename"

Scoring Algorithm

Each rule produces a severity score when a violation is detected:

Severity	Score Range	Description
`low`	0.1 – 0.3	Minor concern — may not require action
`medium`	0.4 – 0.6	Moderate concern — review recommended
`high`	0.7 – 0.9	Significant violation — action likely required
`critical`	0.9 – 1.0	Severe violation — immediate action recommended

The overallScore in the moderation output is the maximum severity across all detected violations. A score of 0.0 means no violations were found.