Taxonomy Collections

Overview

The taxonomy system classifies creator content using 6 super-concept dimensions with 3,208 pre-classified tags. Two collections power this: taxonomy_dimensions for the classification schema, and taxonomy_mapping for tag assignments.

taxonomy_dimensions

Defines the 6 super-concept dimensions used for content classification.

Purpose

Define classification schema for content tagging
Store dimension names, descriptions, and allowed values
Enable AI-powered content discovery and recommendations
Support multi-dimensional content search

The 6 Dimensions

Dimension	Description	Example Values
`body_type`	Physical appearance and build	`athletic`, `curvy`, `petite`, `muscular`, `slim`
`performance_style`	Content tone and presentation	`sensual`, `playful`, `dominant`, `submissive`, `artistic`
`setting`	Location and environment	`bedroom`, `outdoor`, `bathroom`, `office`, `studio`
`attire`	Clothing and accessories	`lingerie`, `casual`, `cosplay`, `fetish_wear`, `nude`
`audience_appeal`	Target audience and niche	`mainstream`, `fetish`, `couples`, `solo`, `group`
`production_quality`	Technical production level	`professional`, `amateur`, `selfie`, `cinematic`

Key Fields

Field	Type	Description
`id`	UUID	Primary key
`dimension_name`	String	Unique dimension identifier (slug)
`display_name`	String	Human-readable name
`description`	Text	Dimension purpose and usage
`allowed_values`	JSON	Array of valid tag values for this dimension
`is_multi_select`	Boolean	Whether multiple values can be selected
`sort_order`	Integer	Display order in UI
`created_at`	DateTime	Dimension creation timestamp

Example Queries

List All Dimensions

await use_mcp_tool({
  server_name: "directus",
  tool_name: "read-items",
  arguments: {
    collection: "taxonomy_dimensions",
    fields: ["id", "dimension_name", "display_name", "description"],
    sort: ["sort_order"]
  }
});

curl -X GET "https://geniehelper.com/api/directus/items/taxonomy_dimensions" \
  -H "Authorization: Bearer YOUR_TOKEN" \
  -G \
  --data-urlencode 'sort=sort_order'

Get Dimension with Values

await use_mcp_tool({
  server_name: "directus",
  tool_name: "read-item",
  arguments: {
    collection: "taxonomy_dimensions",
    id: "dimension-uuid",
    fields: ["dimension_name", "display_name", "allowed_values", "is_multi_select"]
  }
});

curl -X GET "https://geniehelper.com/api/directus/items/taxonomy_dimensions/DIMENSION_ID" \
  -H "Authorization: Bearer YOUR_TOKEN" \
  -G \
  --data-urlencode 'fields=dimension_name,display_name,allowed_values,is_multi_select'

taxonomy_mapping

Stores 3,208 pre-classified tags mapped across the 6 dimensions.

Purpose

Map platform-specific tags to taxonomy dimensions
Enable fast AI-powered content classification
Support content search and discovery
Provide training data for taxonomy classifiers

Classification System

Each tag can have assignments across multiple dimensions:

// Example: "yoga pants" tag mapping
{
  "tag": "yoga pants",
  "platform_source": "onlyfans",
  "dimension_mappings": {
    "attire": ["activewear", "casual"],
    "body_type": ["athletic", "fit"],
    "setting": ["gym", "home"]
  },
  "confidence_score": 0.92
}

Key Fields

Field	Type	Description
`id`	UUID	Primary key
`tag`	String	Original tag text (normalized lowercase)
`platform_source`	String	Platform where tag originated
`dimension_mappings`	JSON	Object mapping dimension names to value arrays
`confidence_score`	Decimal	Classification confidence (0.0-1.0)
`usage_count`	Integer	How many times this tag has been applied
`created_at`	DateTime	Mapping creation timestamp
`updated_at`	DateTime	Last classification update

Example Queries

Search for Tag

await use_mcp_tool({
  server_name: "directus",
  tool_name: "search-items",
  arguments: {
    collection: "taxonomy_mapping",
    query: "lingerie",
    fields: ["tag", "dimension_mappings", "confidence_score"]
  }
});

curl -X GET "https://geniehelper.com/api/directus/items/taxonomy_mapping" \
  -H "Authorization: Bearer YOUR_TOKEN" \
  -G \
  --data-urlencode 'search=lingerie'

Get Tags by Dimension

// Find all tags with "athletic" in body_type dimension
await use_mcp_tool({
  server_name: "directus",
  tool_name: "read-items",
  arguments: {
    collection: "taxonomy_mapping",
    fields: ["tag", "dimension_mappings", "usage_count"],
    filter: {
      dimension_mappings: { _contains: "athletic" }
    },
    sort: ["-usage_count"],
    limit: 50
  }
});

curl -X GET "https://geniehelper.com/api/directus/items/taxonomy_mapping" \
  -H "Authorization: Bearer YOUR_TOKEN" \
  -G \
  --data-urlencode 'filter[dimension_mappings][_contains]=athletic' \
  --data-urlencode 'sort=-usage_count' \
  --data-urlencode 'limit=50'

Create New Mapping

await use_mcp_tool({
  server_name: "directus",
  tool_name: "create-item",
  arguments: {
    collection: "taxonomy_mapping",
    data: {
      tag: "beach photoshoot",
      platform_source: "onlyfans",
      dimension_mappings: {
        setting: ["outdoor", "beach"],
        production_quality: ["professional"],
        performance_style: ["artistic"]
      },
      confidence_score: 0.88
    }
  }
});

curl -X POST "https://geniehelper.com/api/directus/items/taxonomy_mapping" \
  -H "Authorization: Bearer YOUR_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "tag": "beach photoshoot",
    "platform_source": "onlyfans",
    "dimension_mappings": {
      "setting": ["outdoor", "beach"],
      "production_quality": ["professional"],
      "performance_style": ["artistic"]
    },
    "confidence_score": 0.88
  }'

AI Classification

The taxonomy system powers AI-powered content classification:

Action Flow

Use the taxonomy-tag action flow to auto-classify content:

// Agent emits action tag in chat response
"[ACTION:taxonomy-tag:{\"media_id\":\"uuid-here\"}]"

// Action runner executes flow:
// 1. Fetch media item caption + metadata
// 2. Call ollama MCP with scout-fast-tag model
// 3. Match returned tags against taxonomy_mapping
// 4. Apply dimension_mappings to media.taxonomy_tags
// 5. Update scraped_media record

Custom Taxonomy Model

The scout-fast-tag:latest Ollama model is fine-tuned on the 3,208 taxonomy mappings:

// Using ollama MCP to classify content
await use_mcp_tool({
  server_name: "ollama",
  tool_name: "generate",
  arguments: {
    model: "scout-fast-tag:latest",
    prompt: `Classify this content caption into taxonomy tags:\n\n"${caption}"`,
    stream: false
  }
});

scraped_media - Content items tagged with taxonomy classifications
action_flows - The taxonomy-tag flow that applies classifications
agent_audits - Logs of taxonomy classification executions

Workflow Integration

Content Classification Flow

User uploads or scrapes new media with caption
AI chat interface or dashboard triggers classification
Agent emits [ACTION:taxonomy-tag:{"media_id":"..."}]
Action runner fetches media caption + existing tags
Calls Ollama scout-fast-tag model with caption
Model returns relevant tags from 3,208 mapping corpus
System looks up taxonomy_mapping for each tag
Aggregates dimension_mappings into final classification
Updates scraped_media.taxonomy_tags with result
Content now discoverable via dimension-based search

Search and Discovery Flow

User searches for “athletic outdoor content”
System queries taxonomy_mapping for matching tags
Finds tags with body_type: ["athletic"] AND setting: ["outdoor"]
Queries scraped_media where taxonomy_tags contains those values
Returns ranked results by engagement metrics

Best Practices

Normalize tags before lookup - Convert to lowercase, trim whitespace
Use confidence thresholds - Only apply mappings with confidence_score >= 0.7
Track usage - Increment usage_count when applying classifications
Handle multi-select - Check taxonomy_dimensions.is_multi_select before applying
Batch classify - Process multiple media items in single taxonomy flow execution
Update existing classifications - Re-run classification when captions are edited

Data Sources

The 3,208 taxonomy mappings were sourced from:

OnlyFans tag corpus (2,400+ tags)
Fansly category system (600+ tags)
Manual classification of top 200 industry terms
Community-submitted classifications

Model Training

The scout-fast-tag model was fine-tuned using:

Base model: SmolLM 135M parameters
Training data: Nodes/Universe/taxonomy_graph.json (3,205 nodes)
Training script: Nodes/Universe/apply_taxonomy_assignments.py
Fine-tuning method: HITL (Human-in-the-loop) corrections
Accuracy: 87% on validation set

Endpoints

MCP Tools

Collections

Overview

taxonomy_dimensions

Purpose

The 6 Dimensions

Key Fields

Example Queries

List All Dimensions

Get Dimension with Values

taxonomy_mapping

Purpose

Classification System

Key Fields

Example Queries

Search for Tag

Get Tags by Dimension

Create New Mapping

AI Classification

Action Flow

Custom Taxonomy Model

Workflow Integration

Content Classification Flow

Search and Discovery Flow

Best Practices

Data Sources

Model Training

See Also

​Overview

​taxonomy_dimensions

​Purpose

​The 6 Dimensions

​Key Fields

​Example Queries

​List All Dimensions

​Get Dimension with Values

​taxonomy_mapping

​Purpose

​Classification System

​Key Fields

​Example Queries

​Search for Tag

​Get Tags by Dimension

​Create New Mapping

​AI Classification

​Action Flow

​Custom Taxonomy Model

​Related Collections

​Workflow Integration

​Content Classification Flow

​Search and Discovery Flow

​Best Practices

​Data Sources

​Model Training

​See Also

Overview

taxonomy_dimensions

Purpose

The 6 Dimensions

Key Fields

Example Queries

List All Dimensions

Get Dimension with Values

taxonomy_mapping

Purpose

Classification System

Key Fields

Example Queries

Search for Tag

Get Tags by Dimension

Create New Mapping

AI Classification

Action Flow

Custom Taxonomy Model

Related Collections

Workflow Integration

Content Classification Flow

Search and Discovery Flow

Best Practices

Data Sources

Model Training

See Also