Percolators

Percolators in Mixpeek allow developers to search against saved queries, enabling automated real-time content matching.

By leveraging the same syntax as hybrid multimodal searches, percolators save these queries for future use. When indexing an asset, you can specify to percolate the asset across your saved queries. For every match above a supplied relevance threshold, results are appended to a matches collection.

​

Understanding Percolators

​

Key Concepts

• Saved Queries: Predefined search queries stored for reuse
• Hybrid Search: Combining multiple search modalities
• Relevance Threshold: Minimum score for a match to be considered
• Feedback Loop: Incremental improvement of alerts through percolator matches

​

Use Cases

1. Content Monitoring: Automatically detect and flag content that matches specific criteria.
2. Personalized Recommendations: Suggest content based on user preferences and past interactions.
3. Alert Systems: Trigger alerts when new content matches predefined conditions.

​

Code Examples

​

Create Query

POST /percolators/queries
{
  "collection_ids": ["scam"],
  "metadata": {"foo": "bar"},
  "queries": [
    {"vector_index": "multimodal", "value": "https://example.com/image.jpg", "type": "url"},
    {"vector_index": "text", "value": "lorem", "type": "text"}
  ],
  "filters": {
    "AND": [{"key": "metadata.foo", "operator": "eq", "value": "bar"}]
  }
}

This example demonstrates how to create a percolator query. It specifies a collection of queries that include both a URL and text, along with metadata filters.

​

Percolate on Index

POST /index/text
{
  "collection_id": "scam",
  "percolate": {
    "enabled": true,
    "min_relevance": 0.5
  },
  "feature_extractors": {
    "embed": [
      {"field_name": "description", "type": "text", "value": "lorem ipsum", "vector_index": "text"},
      {"field_name": "title", "type": "text", "value": "Thing #1", "vector_index": "keyword"}
    ]
  }
}

This example shows how to percolate an asset when indexing. It enables percolation with a minimum relevance threshold and specifies feature extractors for embedding.

​

Get Matches

POST /percolators/matches
{
  "collection_ids": ["scam"]
}

This example retrieves matches from the percolator for the specified collection. It returns results that meet the criteria defined in the saved queries.