Discovery Search

Discovery search explores the knowledge graph outward from a starting entity to find related entities, uncover hidden connections, and map the neighborhood of a specific data asset. Unlike targeted semantic or structural search, discovery search is exploratory and designed for data catalog browsing.

Overview

Discovery search answers questions like "What else is related to this entity?" and "What does the neighborhood around this dataset look like?" It combines graph expansion with relevance scoring to surface the most interesting related entities.

How It Works

1. Start from a seed entity (identified by URN)
2. Expand outward through relationships up to a configurable depth
3. Score each discovered entity based on relationship distance, entity importance, and access frequency
4. Return the top-k most relevant discovered entities with their relationship paths

API Usage

results = await service.search(SearchQuery(
    query="urn:matih:dataset:acme:customer_events",
    tenant_id="acme",
    mode=SearchMode.DISCOVERY,
    top_k=20,
    filters=SearchFilters(
        max_hops=3,
    ),
))

Discovery Scoring

Discovered entities are scored using a combination of factors:

Relationship Path Tracking

Each discovery result includes the full path from the seed entity:

{
  "entity_urn": "urn:matih:model:acme:churn_predictor",
  "score": 0.85,
  "path": [
    {"from": "urn:matih:dataset:acme:customer_events", "rel": "CONSUMED_BY"},
    {"from": "urn:matih:feature:acme:customer_features", "rel": "TRAINED_ON"},
    {"from": "urn:matih:model:acme:churn_predictor", "rel": null}
  ],
  "hop_distance": 2
}

Filtering Options

Use Cases