📊 Knowledge Graphs Basics

What is a Knowledge Graph?

A Knowledge Graph (KG) is a way of organizing information so that computers can understand relationships between things. Instead of storing facts in text form, a knowledge graph represents information as a network of connected entities and relationships—similar to how your brain organizes concepts.

Think of it this way: If you wanted to tell someone about a disease, you could write a paragraph describing symptoms, treatments, and causes. Or, you could draw a diagram showing the disease node connected to symptom nodes, treatment nodes, and cause nodes. A knowledge graph is exactly that—a visual and logical way to connect concepts.

The Three Components of a Knowledge Graph

1. Entities (Nodes)

Entities are the "things" or "concepts" in your knowledge graph. They represent real-world objects, people, places, or abstract concepts.

2. Relationships (Edges)

Relationships describe how entities are connected. They define the nature of the connection between two entities.

3. Attributes (Properties)

Attributes are properties or characteristics of entities. They describe qualities or measurements associated with an entity.

Visual Example: A Simple Healthcare Knowledge Graph

Here's a simple example of how a knowledge graph looks in a medical context:

In this graph: Circles are entities (Diseases, Drugs, Symptoms). Arrows are relationships describing how entities connect.

Why Are Knowledge Graphs Useful?

Building a Knowledge Graph: Step-by-Step

Step 1: Identify Entities

List all the important concepts or objects in your domain:

Step 2: Define Relationships

Specify how entities connect:

Step 3: Extract or Create Data

Populate your graph with actual data:

{
  "entities": [
    {"id": "D001", "name": "Diabetes", "type": "Disease"},
    {"id": "S001", "name": "Fatigue", "type": "Symptom"},
    {"id": "M001", "name": "Metformin", "type": "Drug"}
  ],
  "relationships": [
    {"source": "D001", "target": "S001", "type": "hasSymptom"},
    {"source": "D001", "target": "M001", "type": "treatedBy"}
  ]
}

Step 4: Implement the Graph

Store it in a graph database like Neo4j or networkx:

from networkx import DiGraph

# Create graph
kg = DiGraph()

# Add entities
kg.add_node("Diabetes", type="Disease")
kg.add_node("Fatigue", type="Symptom")
kg.add_node("Metformin", type="Drug")

# Add relationships
kg.add_edge("Diabetes", "Fatigue", relation="hasSymptom")
kg.add_edge("Diabetes", "Metformin", relation="treatedBy")

# Query: What symptoms does Diabetes have?
symptoms = [target for source, target, data in kg.out_edges(data=True)
            if data['relation'] == 'hasSymptom']

Knowledge Graphs vs. Other Data Structures

Real-World Applications of Knowledge Graphs

Healthcare

Medical knowledge graphs connect diseases, symptoms, drugs, treatments, and patient records. They enable:

• Clinical decision support (what tests to order, what treatments are recommended)
• Drug interaction checking (is this drug safe with the patient's other medications?)
• Patient risk stratification (who is at high risk for complications?)

Google Search

Google's Knowledge Graph powers search results. It connects entities like people, places, movies, and organizations so that search can understand intent. Example: When you search "Einstein," Google shows you his biography, related people, his theories, and more—all from the knowledge graph.

Social Networks

Facebook and LinkedIn use knowledge graphs to represent people, relationships, companies, and interests. This enables friend recommendations, job suggestions, and targeted advertising.

E-Commerce

Amazon and other retailers use knowledge graphs to connect products, categories, attributes, reviews, and customer preferences. This powers recommendations: "Customers who bought X also bought Y."

Knowledge Graph Sizes: From Small to Massive

Key Concepts to Remember

What's Next?

Now that you understand the basics of knowledge graphs, the next chapter will dive deeper into:

• How to query a knowledge graph (SPARQL, Cypher)
• Knowledge graph embeddings (representing entities as vectors)
• Inference and reasoning over a knowledge graph
• Integrating knowledge graphs with neural networks