semantica/docs/reference/semantic_extract.md

title, description, icon

title	description	icon
Semantic Extract Module	Named entity recognition, relation extraction, event detection, and triplet generation.	magnifying-glass-chart

semantica.semantic_extract extracts structured information from unstructured text: the foundation of every knowledge graph in Semantica:

• NERExtractor: named entity recognition with confidence scores and source attribution
• RelationExtractor: typed relationship extraction (founded_by, located_in, and custom types)
• TripletExtractor: direct (subject, predicate, object) triplet generation for RDF output
• EventDetector: event detection with participants, temporal context, and confidence
• Three extraction modes on every extractor: "pattern" (no API key), "huggingface", "llm"

Getting Started

Prerequisites & Setup

Step 1: Install Dependencies

# Basic extraction (pattern methods only)
pip install semantica

# HuggingFace models for advanced NER
pip install semantica[models-huggingface]

# LLM-based extraction (highest accuracy)
pip install semantica[llm-groq]    # or llm-openai

Step 2: Set API Keys (for LLM methods only)

export GROQ_API_KEY="your_groq_key_here"
export OPENAI_API_KEY="your_openai_key_here"

Step 3: First Extraction

from semantica.semantic_extract import NERExtractor

# Start with pattern method (no setup required)
ner = NERExtractor(method="pattern")
entities = ner.extract("Apple Inc. was founded by Steve Jobs.")
print(f"Found {len(entities)} entities")
# Output: Found 2 entities

# Upgrade to LLM for better accuracy
from semantica.llms import Groq
import os

llm = Groq(api_key=os.getenv("GROQ_API_KEY"))
ner = NERExtractor(method="llm", llm_provider=llm)
entities = ner.extract("Apple Inc. was founded by Steve Jobs.")

Exported Classes

**`NamedEntityRecognizer`** is the high-level coordinator with confidence thresholding and overlap merging. **`NERExtractor`** is the lower-level implementation. For most use cases, start with `NERExtractor` for simplicity or `NamedEntityRecognizer` for fine-grained control.

Class	Role
NamedEntityRecognizer	High-level NER with confidence thresholding and overlap merging
NERExtractor	Core NER implementation: use directly for simplicity
RelationExtractor	Typed relationship extraction (founded_by, located_in, ...)
TripletExtractor	Direct (subject, predicate, object) triplet generation for RDF output
EventDetector	Event detection with participants, temporal context, and confidence scores
CoreferenceResolver	Resolve "Apple" and "the company" to the same canonical entity
Entity	{id, text, type, confidence, start, end}
Relation	{subject, predicate, object, confidence}
Event	{type, participants, temporal, location, confidence}

Method Selection Guide

Zero dependencies, no API key required. Uses spaCy rules and regex to match standard entity types.

| | |
| :-- | :-- |
| **Setup** | None: works out of the box |
| **Cost** | Free |
| **Accuracy** | Good for standard entity types |
| **Best for** | Quick prototyping, batch processing, air-gapped systems |

```python
from semantica.semantic_extract import NERExtractor, RelationExtractor

ner = NERExtractor(method="pattern")
entities = ner.extract("Apple Inc. was founded by Steve Jobs in Cupertino.")

rel = RelationExtractor(method="pattern")
relationships = rel.extract(text, entities=entities)
```

Use any pre-trained or fine-tuned transformer model. Free inference, runs locally.

| | |
| :-- | :-- |
| **Setup** | `pip install semantica[models-huggingface]` |
| **Cost** | Free (local compute) |
| **Accuracy** | Excellent for domain-specific NER |
| **Best for** | Medical NER, custom fine-tunes, no API cost |

```python
from semantica.semantic_extract import NERExtractor

ner = NERExtractor(method="huggingface")

# Pass model per-call
entities = ner.extract(text, model="dslim/bert-base-NER", device="cpu")

# Biomedical NER
entities = ner.extract(text, model="d4data/biomedical-ner-all")
```

Highest accuracy for complex schemas and custom entity types. Requires an LLM API key.

| | |
| :-- | :-- |
| **Setup** | `pip install semantica[llm-groq]` + API key |
| **Cost** | Depends on provider |
| **Accuracy** | Highest: handles complex types and context |
| **Best for** | Production, custom entity types, complex relation schemas |

```python
import os
from semantica.llms import Groq
from semantica.semantic_extract import NERExtractor, RelationExtractor, TripletExtractor

llm = Groq(model="llama-3.3-70b-versatile", api_key=os.getenv("GROQ_API_KEY"))

ner  = NERExtractor(method="llm",  llm_provider=llm, max_retries=3)
rel  = RelationExtractor(method="llm", llm_provider=llm)
trip = TripletExtractor(method="llm", llm_provider=llm)

entities      = ner.extract(text)
relationships = rel.extract(text, entities=entities)
triplets      = trip.extract(text)
```

Try methods in priority order: guarantees non-empty results even when the preferred method is unavailable.

```python
from semantica.semantic_extract import NERExtractor, RelationExtractor

# Try LLM first, fall back to pattern on error
ner = NERExtractor(method=["llm", "pattern"])
rel = RelationExtractor(method=["llm", "pattern"])

# Always returns results: safe for production pipelines
entities      = ner.extract(text)
relationships = rel.extract(text, entities=entities)
```

<Tip>
  Use fallback chains in pipelines where API availability isn't guaranteed (rate limits, network issues). The first method in the list is always tried first.
</Tip>

Method Availability by Extractor

Extractor	pattern	huggingface	llm	Notes
NERExtractor	✅	✅	✅	Full method support
RelationExtractor	✅	✅	✅	Also supports dependency, cooccurrence
TripletExtractor	✅	✅	✅	Also supports rules method
EventDetector	✅	❌	✅	Pattern and LLM only

Method Fallback Chains

For reliability, extractors support fallback chains that try methods in order until one succeeds:

# Try LLM first, fall back to pattern if it fails
ner = NERExtractor(method=["llm", "pattern"])
rel = RelationExtractor(method=["llm", "pattern"]) 
trip = TripletExtractor(method=["llm", "pattern"])

# Always returns results - guarantees non-empty extraction
entities = ner.extract(text)

NER Merge Strategies

NERExtractor uses merge_strategy="fallback" by default, so a method list remains an ordered fallback chain. To run several methods together, choose one of the explicit strategies below:

Strategy	Behavior
fallback	Return the first non-empty method result.
union	Keep candidates from any method. Same-label boundary variants are aligned, while distinct labels remain available.
consensus	Require cross-method support for an offset-aligned candidate. min_votes defaults to 2.

from semantica.semantic_extract import NERExtractor

ner = NERExtractor(
    method=["spacy", "huggingface"],
    merge_strategy="consensus",
    min_votes=2,
    min_agreement=0.75,  # optional support-ratio requirement
    method_weights={"spacy": 0.8, "huggingface": 1.0},
)
entities = ner.extract(text)

for entity in entities:
    print(entity.metadata["supporting_methods"])
    print(entity.metadata["vote_count"], entity.metadata["agreement"])

Consensus counts support against the configured eligible methods, not only methods that emitted a candidate. An empty or failed eligible method is therefore a non-supporting vote. Use eligible_methods=[...] to restrict the consensus denominator when the configured methods have different coverage, or use merge_strategy="union" for complementary rule extractors. method_weights only break an otherwise eligible exact-span cross-label tie; they never turn one method into multiple votes.

Each merged entity includes supporting_methods, vote_count, eligible_method_count, agreement, and per-method method_scores in its metadata. Consensus treats compatible label aliases such as PER/PERSON and ORGANIZATION/ORG as the same vote. It resolves a cross-label conflict only when the final spans are identical, using method weight, vote count, confidence, and a stable label order; nested entities at different spans remain available. ml and spacy are one backend for both voting and weights, so their weights are interchangeable (conflicting values are rejected). Boundary candidates are matched one-to-one only when their span IoU is at least 0.5 with every existing vote in that candidate; equal-confidence variants prefer the longer span. If a provider omits offsets, Semantica resolves its entity text against whole-word document matches before merging. This keeps repeated mentions with the same text distinct and prevents one broad span from acting as a vote for multiple mentions.

ensemble_voting=True is deprecated and maps to merge_strategy="union" during migration. Use merge_strategy="consensus" when method agreement is required.

Unlike fallback, union and consensus never inject a pattern-derived entity after the configured methods return no candidates. An empty result is therefore meaningful in those strategies.

Quick Start

from semantica.semantic_extract import NERExtractor, RelationExtractor, TripletExtractor
from semantica.llms import Groq
import os

text = "Apple Inc. was founded by Steve Jobs in Cupertino in 1976."
llm  = Groq(model="llama-3.3-70b-versatile", api_key=os.getenv("GROQ_API_KEY"))

entities      = NERExtractor(method="llm", llm_provider=llm).extract(text)
relationships = RelationExtractor(method="llm", llm_provider=llm).extract(text, entities=entities)
triplets      = TripletExtractor(method="llm", llm_provider=llm).extract(text)

<img src="/assets/img/diagrams/extraction-pipeline.svg" alt="Semantic extraction pipeline: raw text fans into NER, Relation, and Coreference extractors, then merges into a Triplet Generator" style={{ width: '100%', borderRadius: '12px', margin: '0 0 24px' }} />

Extractor Methods

Method	Returns	Description
extract(text)	List[Entity] / List[Relation] / List[Triplet] / List[Event]	Extract from single text input
extract(texts)	List[List[...]]	Process multiple texts (batch detected automatically)

NERExtractor

from semantica.semantic_extract import NERExtractor
from semantica.llms import Groq
import os

# Pattern-based: fast, no API key, good for standard entity types
ner = NERExtractor(method="pattern")
entities = ner.extract("Apple Inc. was founded by Steve Jobs in Cupertino.")

# HuggingFace-based: custom models, no API cost
ner = NERExtractor(method="huggingface")
entities = ner.extract(text, model="dslim/bert-base-NER", device="cpu")

# LLM-based: best accuracy, handles complex schemas and custom types
llm = Groq(model="llama-3.3-70b-versatile", api_key=os.getenv("GROQ_API_KEY"))
ner = NERExtractor(method="llm", llm_provider=llm, max_retries=3)
entities = ner.extract(text)

Output format:

[
    {"text": "Apple Inc.",  "type": "ORGANIZATION", "confidence": 0.98, "start": 0,  "end": 10},
    {"text": "Steve Jobs",  "type": "PERSON",       "confidence": 0.99, "start": 27, "end": 37},
    {"text": "Cupertino",   "type": "LOCATION",     "confidence": 0.97, "start": 41, "end": 50}
]

Custom Entity Types

ner = NERExtractor(
    method="pattern",
    custom_entities={
        "DRUG": ["aspirin", "ibuprofen", "metformin"],
        "GENE": ["BRCA1", "TP53", "EGFR"]
    }
)

**v0.5.0 fix:** `NERExtractor(method="llm")` no longer silently falls back to pattern extraction on custom gateways. The `response_format=json_object` parameter is now conditionally omitted for incompatible gateways, with a plain `generate()` + JSON parsing fallback applied automatically.

RelationExtractor

from semantica.semantic_extract import RelationExtractor

rel = RelationExtractor(method="llm", llm_provider=llm, max_retries=3)
relationships = rel.extract(text, entities=entities)

Output format:

[
    {"subject": "Steve Jobs", "predicate": "founded",    "object": "Apple Inc.", "confidence": 0.92},
    {"subject": "Apple Inc.", "predicate": "located_in", "object": "Cupertino",  "confidence": 0.89}
]

Available methods:

• "pattern": rule-based pattern matching
• "dependency": spaCy dependency parsing
• "cooccurrence": proximity-based co-occurrence
• "huggingface": custom models
• "llm": highest accuracy, requires API key

TripletExtractor

Generate RDF-ready (subject, predicate, object) triplets directly from text:

from semantica.semantic_extract import TripletExtractor

trip = TripletExtractor(method="llm", llm_provider=llm)
triplets = trip.extract(text)
# → [{"subject": "Steve Jobs", "predicate": "founded", "object": "Apple Inc.", ...}]

Triplets are suitable for loading directly into a triplet store or knowledge graph.

EventDetector

Detect events with participants and temporal context:

from typing import List
from semantica.semantic_extract import EventDetector, Event

extractor = EventDetector(method="llm", llm_provider=llm)
events: List[Event] = extractor.extract(text)

for event in events:
    print(f"Event type:   {event.type}")
    print(f"Participants: {event.participants}")
    print(f"Temporal:     {event.temporal}")
    print(f"Confidence:   {event.confidence:.2f}")

Output fields per event:

• type: event category (e.g. "founding", "acquisition")
• participants: list of entities with roles
• temporal: date or time reference
• location: location entity (when present)
• confidence: extraction confidence score

CoreferenceResolver

Resolve pronoun and alias references to canonical entities before extraction:

from semantica.semantic_extract import CoreferenceResolver

resolver = CoreferenceResolver()
resolved_text = resolver.resolve(
    "Apple Inc. was founded in 1976. The company is headquartered in Cupertino."
)
# "Apple Inc." replaces "The company" for consistent downstream extraction

Batch Processing

All extractors automatically detect batch input and process multiple texts efficiently:

# Batch processing with list input
texts = ["Apple Inc. was founded by Steve Jobs.", "Google was founded by Larry Page.", "Microsoft was founded by Bill Gates."]

ner = NERExtractor(method="llm", llm_provider=llm)
batch_results = ner.extract(texts)  # Returns List[List[Entity]]

# Process results
for i, doc_entities in enumerate(batch_results):
    print(f"Document {i}: {len(doc_entities)} entities")
    for entity in doc_entities:
        print(f"  - {entity.text} ({entity.label})")

Batch Input Options:

# Option 1: List of strings
texts = ["Text 1...", "Text 2...", "Text 3..."]
results = ner.extract(texts)

# Option 2: List of documents with IDs (adds provenance metadata)
documents = [
    {"id": "doc_1", "content": "Apple Inc. was founded by Steve Jobs."},
    {"id": "doc_2", "content": "Google was founded by Larry Page."}
]
results = ner.extract(documents)  # Entities include document_id in metadata

Using All Extractors Together

The standard extraction pipeline: entities → relationships → triplets:

from semantica.semantic_extract import NERExtractor, RelationExtractor, TripletExtractor
from semantica.llms import Groq
import os

llm = Groq(model="llama-3.3-70b-versatile", api_key=os.getenv("GROQ_API_KEY"))

ner  = NERExtractor(method="llm",      llm_provider=llm, max_retries=3)
rel  = RelationExtractor(method="llm", llm_provider=llm, max_retries=3)
trip = TripletExtractor(method="llm",  llm_provider=llm, max_retries=3)

entities      = ner.extract(text)
relationships = rel.extract(text, entities=entities)
triplets      = trip.extract(text)

Extraction Method Comparison

Method	Speed	Cost	Accuracy	Custom Types
pattern	Very fast	Free	Medium	Yes (dictionary)
ml	Fast	Free	High	Limited
llm	Medium	API cost	Highest	Yes (schema)

• LLM Providers — Configure which LLM is used for extraction.
• Knowledge Graph — Build graphs from extracted entities and relationships.
• Parse Module — Parse documents before extraction.
• Deduplication — Resolve duplicate entities after extraction.