Building Knowledge Graphs from Unstructured Text with Claude

You have a pile of unstructured documents and need to answer questions that span them — "who works with people who worked on project X", "which vendors are connected to this incident". No single document contains the answer. RAG retrieval won't chain the facts for you. You need a knowledge graph: entities as nodes, typed relations as edges, so that multi-hop reasoning becomes graph traversal.

Building one used to mean training a named-entity recognizer on your domain, training a relation classifier, writing entity-resolution heuristics, and maintaining all three as your data shifted. With Claude, each of those stages becomes a prompt.

What You'll Learn

By the end of this guide you will be able to:

• Use structured outputs to extract typed entities and subject–predicate–object triples from arbitrary text with no training data
• Apply Claude-driven entity resolution to collapse surface-form variants into canonical nodes, replacing brittle string-similarity heuristics
• Assemble and query an in-memory graph, and run multi-hop questions by serializing subgraphs back to Claude
• Measure extraction quality with precision/recall against a gold set and reason about the cost/quality tradeoff between Haiku and Sonnet

Everything runs in memory with no database. The techniques transfer directly to Neo4j, Neptune, or a Postgres adjacency table when you need to scale.

Prerequisites

• Python 3.11+
• Anthropic API key (get one here)
• Basic familiarity with graphs (nodes, edges, traversal)

Setup

We use two models. Haiku handles the high-volume, schema-constrained extraction work where speed and cost matter more than nuance. Sonnet handles entity resolution and summarization, where the model needs to weigh conflicting evidence across documents.

import anthropic
from pydantic import BaseModel, Field
from typing import List, Optional
client = anthropic.Anthropic()
Define your extraction schema
class Entity(BaseModel):
    name: str = Field(description="The canonical name of the entity")
    type: str = Field(description="Entity type: PERSON, ORG, LOC, EVENT, etc.")
    description: str = Field(description="One-line context for disambiguation")
class Relation(BaseModel):
    subject: str = Field(description="Name of the subject entity")
    predicate: str = Field(description="Relation type in present tense")
    object: str = Field(description="Name of the object entity")
class Extraction(BaseModel):
    entities: List[Entity]
    relations: List[Relation]

Building a Corpus

We need a handful of documents that talk about overlapping entities, so that entity resolution has real work to do. The Apollo program is a good test bed: six short Wikipedia summaries that all mention NASA, the Moon, several astronauts, and a launch vehicle — but each article names them slightly differently.

We fetch summaries from the Wikipedia REST API rather than full articles to keep token costs low. For a production pipeline you would chunk full documents; the extraction logic is identical.

Entity and Relation Extraction

Classical NER tags spans of text with labels (PERSON, ORG, LOC). Classical relation extraction then classifies pairs of spans into relation types. Both traditionally require labeled training data per domain.

We collapse both stages into a single Claude call per document. The key is structured outputs: we define the output shape as a Pydantic model and pass it to client.messages.parse(). Claude's response is guaranteed to validate against that schema and comes back as a typed Python object — no regex parsing, no JSON decode errors, no defensive isinstance checks.

def extract_from_document(text: str) -> Extraction:
    response = client.messages.parse(
        model="claude-3-haiku-20240307",
        max_tokens=2000,
        system="Extract all named entities and their relationships from the text. "
               "Be thorough but precise.",
        messages=[{"role": "user", "content": text}],
        response_model=Extraction
    )
    return response

Let's look at what was extracted. Notice how the same real-world entity appears under different surface forms across documents — this is the entity resolution problem we solve next.

Entity Resolution

The raw extraction gives us overlapping mentions: "NASA" and "National Aeronautics and Space Administration", "Neil Armstrong" and "Armstrong", possibly "the Moon" and "Moon". If we build a graph directly from this, we get a fractured mess where the same concept is split across disconnected nodes.

Traditional approaches use string similarity (edit distance, Jaccard on tokens) plus blocking rules. That works for typos but fails on "Edwin Aldrin" vs "Buzz Aldrin" — two names with zero character overlap that refer to the same person.

We instead ask Claude to cluster entities of each type, using the one-line descriptions from extraction as disambiguation context. The descriptions matter: "Armstrong — first person to walk on the Moon" and "Armstrong — jazz trumpeter" have the same name but should not merge.

def resolve_entities(entities: List[Entity]) -> dict:
    """Returns a mapping from alias -> canonical name"""
    # Group by entity type
    by_type = {}
    for e in entities:
        by_type.setdefault(e.type, []).append(e)
    
    alias_map = {}
    for etype, group in by_type.items():
        prompt = f"""Cluster these {etype} entities that refer to the same real-world thing.
        For each cluster, pick the best canonical name.
        
        Entities:
        {chr(10).join(f'- {e.name}: {e.description}' for e in group)}
        
        Return as JSON: {{clusters: [["name1", "name2"], ...], canonical: {{"name": "canonical"}}}}
        """
        response = client.messages.parse(
            model="claude-3-sonnet-20240229",
            max_tokens=1000,
            messages=[{"role": "user", "content": prompt}],
            response_model=ResolutionOutput
        )
        for cluster in response.clusters:
            canonical = response.canonical[cluster[0]]
            for alias in cluster:
                alias_map[alias] = canonical
    return alias_map

Two failure modes to watch for. First, any raw name Claude leaves out of every cluster silently disappears from the graph, because alias_to_canonical has no entry for it — a production resolver should fall back to a single-element cluster for unmatched names so nothing is lost. Second, the resolver can over-merge: a specific mission like "Gemini 12" may get folded into the broader "Project Gemini" because the descriptions overlap. The first loses nodes, the second loses precision. Both are worth spot-checking in the output below.

Assembling the Graph

With a clean alias map, we rewrite every relation endpoint to its canonical form and load the result into NetworkX. We use a MultiDiGraph because two entities can be connected by several distinct predicates ("launched from" and "operated by"), and direction matters ("Armstrong commanded Apollo 11" is not the same edge as "Apollo 11 commanded Armstrong").

Each node carries its type, the source document, and the original surface form for auditability.

import networkx as nx
def build_graph(extractions: List[Extraction], alias_map: dict) -> nx.MultiDiGraph:
    G = nx.MultiDiGraph()
    
    for extraction in extractions:
        for entity in extraction.entities:
            canonical = alias_map.get(entity.name, entity.name)
            G.add_node(canonical, type=entity.type, 
                      description=entity.description)
        
        for relation in extraction.relations:
            subj = alias_map.get(relation.subject, relation.subject)
            obj = alias_map.get(relation.object, relation.object)
            G.add_edge(subj, obj, predicate=relation.predicate)
    
    return G

Querying the Graph with Multi-Hop Reasoning

Now the real power: answering questions that require chaining facts across documents. We serialize the relevant subgraph back to Claude for reasoning.

def query_graph(G: nx.MultiDiGraph, question: str) -> str:
    # Serialize a relevant subgraph (simplified - in practice, use BFS from seed nodes)
    subgraph_text = ""
    for node, data in G.nodes(data=True):
        subgraph_text += f"{node} ({data['type']}): {data.get('description', '')}\n"
    for u, v, data in G.edges(data=True):
        subgraph_text += f"{u} --[{data['predicate']}]--> {v}\n"
    
    prompt = f"""Given this knowledge graph:
    {subgraph_text}
    
    Answer: {question}
    """
    
    response = client.messages.create(
        model="claude-3-sonnet-20240229",
        max_tokens=500,
        messages=[{"role": "user", "content": prompt}]
    )
    return response.content[0].text

Measuring Quality

To trust your graph in production, you need to measure precision and recall against a gold standard. Create a small annotated set of documents with known entities and relations, then compare Claude's extraction against it.

def evaluate_extraction(gold: Extraction, predicted: Extraction):
    gold_entities = set((e.name, e.type) for e in gold.entities)
    pred_entities = set((e.name, e.type) for e in predicted.entities)
    
    true_positives = gold_entities & pred_entities
    false_positives = pred_entities - gold_entities
    false_negatives = gold_entities - pred_entities
    
    precision = len(true_positives) / len(pred_entities) if pred_entities else 0
    recall = len(true_positives) / len(gold_entities) if gold_entities else 0
    f1 = 2  precision  recall / (precision + recall) if (precision + recall) else 0
    
    return {"precision": precision, "recall": recall, "f1": f1}

Cost/Quality Tradeoffs

• Haiku ($0.25/M input tokens): Best for high-volume extraction where schema is well-defined. Expect slightly lower recall on rare entity types.
• Sonnet ($3.00/M input tokens): Use for entity resolution and complex reasoning. Better at handling ambiguous references and subtle distinctions.

A common pattern: extract with Haiku, resolve with Sonnet. This gives you 80% of the quality at 30% of the cost.

Key Takeaways

• No training data needed: Claude's structured outputs let you define entity and relation schemas with Pydantic models, eliminating the need for domain-specific NER/relation classifiers.
• AI-driven entity resolution beats heuristics: Claude can resolve "Buzz Aldrin" vs "Edwin Aldrin" using context, where string similarity would fail completely.
• Multi-hop reasoning becomes graph traversal: Once your knowledge graph is built, complex questions that span documents become simple path-finding problems.
• Start with an in-memory graph: NetworkX is perfect for prototyping. The same extraction and resolution logic transfers directly to production graph databases.
• Measure and iterate: Always evaluate precision/recall against a gold set, and watch for the two failure modes: dropped entities and over-merging.