Blog 1: How the World Thinks in Graphs: From Reality to Relational Intelligence

We are not just beings with knowledge. We are beings of relationships,with people, ideas, places, and time.

Whether you're Googling "best vegan recipes," diagnosing rare diseases, or tracing how an idea sparked across academia, you're dealing with networks of knowledge,not isolated facts. And the smartest AI systems today do the same: they think in graphs.

What Is a Knowledge Graph?

Imagine walking into a hospital: A Doctor treats a Patient. This Patient has_disease Diabetes, a condition that is treated_by Insulin. The Insulin, in turn, interacts_with a specific Receptor in the body. Each triple represents a directed edge in a graph a semantic relationship between two entities. That’s a Knowledge Graph.

Formally: A KG is a set of triples \( (h, r, t) \) where:

• \( h \): head entity
• \( r \): relation
• \( t \): tail entity

Why Graphs, Not Tables?

Traditional databases work in rows and columns. But real life? It’s messy, contextual, and interconnected. Graphs make those connections explicit , no JOINs, just walks. Want to find: "Which drugs interact with the medications prescribed to people who have diabetes and hypertension?"

In a KG, this is just a walk through nodes. In SQL? You’ll need five JOINs and some aspirin.

Real-World Applications of Knowledge Graphs

• Google Knowledge Graph: Summarizes 500B+ facts about the world.
• Drug Discovery: BioKG, AlphaFold use graphs to understand proteins and diseases.
• Fraud Detection: Banks track transaction networks.
• Content Recommendations: Netflix and Spotify map media relationships.

Types of Knowledge Graphs

\[ \begin{array}{|c|c|c|} \hline \textbf{Type} & \textbf{Example} & \textbf{Feature} \\ \hline \text{Static} & \text{Wikipedia, DBpedia} & \text{Manually curated, rarely updated} \\ \hline \text{Dynamic} & \text{Social media graphs} & \text{Changes in real-time} \\ \hline \text{Multimodal} & \text{OpenAI GPT-4o + KG} & \text{Combines text, images, audio} \\ \hline \text{Temporal} & \text{News graphs} & \text{Adds time to edges} \\ \hline \end{array} \]

But Here's the Problem...

ML models need numbers, not text-based triples. We can’t feed \((\text{Insulin}, \text{treats}, \text{Diabetes})\) directly into a neural network.

So how do we convert relational knowledge into a computable format?

Coming Up Next

That’s where Knowledge Graph Embeddings come in. They embed entities and relations into vector spaces, capturing structure and semantics.

In the next blog, we’ll explore:

• What embeddings are
• How models like TransE work
• How to predict missing links using learned vectors

TL;DR

• The world is relational. Graphs capture that.
• Knowledge graphs model entities and their connections.
• They power real-world applications from Google to genomics.
• But to use them in ML, we must embed them numerically.

Quote to Leave You With:

“A graph is not just data , it’s a story. A trail of ‘whys’ and ‘hows’ that even humans find hard to untangle. And AI? It’s just learning to walk those paths.”