Status: Draft. This is a dataset project, not a competition.

Objective

Segment UK e-commerce customers into behavioural profiles from their purchases.

Approach

1. MiniLM embeddings of item descriptions, so products that are similar but worded differently end up close together.
2. K-Means, choosing k with silhouette scoring.
3. Per-customer aggregates of what they bought, how often, and how much.
4. Self-organising map (planned), to see how the segments sit next to each other rather than only which one a customer falls in.

Where it's at

• ✅ Item embeddings and K-Means clustering
• ✅ Per-customer aggregates
• ❌ Self-organising map