semantica/docs/reference/seed.md

Seed

Seed data management system for initializing Knowledge Graphs from verified sources.

---

🎯 Overview

• :material-database-import:{ .lg .middle } Multi-Source Loading

  ---

  Load seed data from CSV, JSON, Databases, and APIs

• :material-graph-outline:{ .lg .middle } Foundation Graph

  ---

  Create reliable foundation graphs to bootstrap your KG

• :material-merge:{ .lg .middle } Data Integration

  ---

  Merge seed data with extracted data using configurable strategies

• :material-check-all:{ .lg .middle } Validation

  ---

  Validate seed data quality and schema compliance

• :material-git:{ .lg .middle } Versioning

  ---

  Manage versions of seed data sources

• :material-export:{ .lg .middle } Export

  ---

  Export seed data to standard formats

!!! tip "When to Use" - Bootstrapping: When starting a new KG and you have existing structured data (taxonomies, user lists, product catalogs). - Reference Data: To load immutable reference data (countries, codes, constants). - Testing: To load consistent test datasets for development.

---

⚙️ Algorithms Used

Data Loading

• Format Detection: Auto-detection of CSV delimiters, JSON structure
• Streaming: Row-by-row processing for large files
• Normalization: Type conversion and encoding handling

Integration & Merging

• Seed-First Strategy: Seed data overrides extracted data (Trust Seed)
• Extracted-First Strategy: Extracted data overrides seed (Trust Extraction)
• Smart Merge: Property-level merging with conflict resolution
• ID Matching: Entity resolution between seed and extracted entities

Validation

• Schema Validation: Template-based structure checking
• Constraint Checking: Required field and type validation
• Consistency Check: Reference integrity (relationships point to existing entities)

---

Main Classes

SeedDataManager

Coordinator for all seed data operations.

Methods:

Method	Description
register_source(name, format, location)	Add data source
create_foundation_graph()	Build KG from sources
validate_quality(seed_data)	Check data quality
integrate_with_extracted(seed, extracted)	Merge graphs
export_seed_data(path, format)	Export data

Example:

from semantica.seed import SeedDataManager

manager = SeedDataManager()
manager.register_source("countries", "csv", "data/countries.csv")
foundation_kg = manager.create_foundation_graph()

SeedDataSource

Data class defining a source.

Attributes:

• name: Source identifier
• type: csv, json, api, sql
• path: File path or connection string
• config: Parsing options

---

Configuration

Environment Variables

export SEED_DATA_DIR=./data/seed
export SEED_MERGE_STRATEGY=seed_first

YAML Configuration

seed:
  sources:
    - name: "employees"
      type: "csv"
      path: "./data/employees.csv"
      
  merge:
    strategy: "seed_first"
    
  validation:
    strict: true

---

Integration Examples

Bootstrapping a KG

from semantica.seed import SeedDataManager
from semantica.ingest import Ingestor
from semantica.kg import KnowledgeGraph

# 1. Load Foundation (Seed)
seed_manager = SeedDataManager()
seed_manager.register_source("taxonomy", "json", "taxonomy.json")
foundation_kg = seed_manager.create_foundation_graph()

# 2. Ingest New Data
ingestor = Ingestor()
new_data = ingestor.ingest("news_articles.pdf")

# 3. Merge
final_kg = seed_manager.integrate_seed_extracted(
    seed_graph=foundation_kg,
    extracted_data=new_data,
    strategy="seed_first"  # Keep taxonomy strict
)

---

Best Practices

1. Trust Seed Data: Usually, seed data is verified. Use seed_first merge strategy.
2. Version Control: Keep seed data files in version control (git).
3. Validate Schema: Ensure seed data matches your target ontology.
4. Clean IDs: Use consistent ID schemes in seed data to facilitate merging.

---

See Also

• Ingest Module - Loading unstructured data
• Knowledge Graph Module - The target graph structure
• Deduplication Module - Handling duplicates during merge