Designing a Knowledge Graph System

A knowledge graph represents real-world entities and the relationships between them, enabling semantic queries and inference.

What is a Knowledge Graph?

Entities (Nodes):
  Person: "Alan Turing"
  Place:  "Cambridge"
  Org:    "University of Cambridge"

Relationships (Edges):
  Alan Turing --[studied_at]--> University of Cambridge
  Alan Turing --[born_in]--> London
  Alan Turing --[invented]--> Turing Machine

RDF Triples

Knowledge represented as subject-predicate-object triples:

<alan_turing> <studied_at> <cambridge_university> .
<alan_turing> <born_in>    <london> .
<alan_turing> <born_on>    "1912-06-23"^^xsd:date .
<alan_turing> <type>       <Person> .

SPARQL Query Language

# Find all people who studied at Cambridge
SELECT ?person ?name WHERE {
  ?person rdf:type :Person .
  ?person :studied_at :cambridge_university .
  ?person :name ?name .
}

# Find colleagues (studied same place)
SELECT ?colleague WHERE {
  :alan_turing :studied_at ?place .
  ?colleague   :studied_at ?place .
  FILTER(?colleague != :alan_turing)
}

Entity Linking (NLP → Graph)

Text: "Turing worked at Bletchley Park during WWII"
    ↓
NER: Turing [PERSON], Bletchley Park [LOCATION], WWII [EVENT]
    ↓
Entity linking: Turing → :alan_turing (confidence: 0.98)
    ↓
Relation extraction: worked_at(alan_turing, bletchley_park)
    ↓
Insert triple into knowledge graph

Inference / Reasoning

Known facts:
  Alan Turing is a Person
  Person is a subclass of Human

Inferred:
  Alan Turing is a Human (via OWL reasoning)

Semantic Search

Query: "Who invented the computer?"
    ↓
Entity extraction + query graph
    ↓
SPARQL: ?person :invented ?thing WHERE ?thing :type :Computer
    ↓
Rank by confidence + relevance
    ↓
Return: Charles Babbage, Alan Turing, John von Neumann

Storage

RDF Store (Triplestore):
  Apache Jena, Amazon Neptune, Stardog
  Optimized for triple patterns

Property Graph:
  Neo4j, TigerGraph
  More natural for software engineers

Conclusion

Knowledge graphs power semantic search, recommendation systems, and question answering. RDF triples + SPARQL + ontologies form the standards layer, while property graph databases offer developer-friendly alternatives.