New AI Groups Products by Meaning to Improve Online Shopping
Could save you time hunting for products and uncover hidden gems you'd miss.
When you shop on a big marketplace, you might see dozens of listings for what is essentially the same thing: a black t-shirt, a black tee, and a men's black shirt. Humans know these are the same, but computers often don't. That's why stores use rigid, pre-made categories like "apparel" which are too broad to capture that one is a fitted tee and another is a loose tank. This paper tackles that problem with a "product family tree" built automatically from how products actually look and are described.
The system, called a Semantic ID, learns from product details to organize everything into meaningful groups, from big categories down to very specific types. Think of it like a smart library that not only shelves books by genre but also groups "cozy mysteries with cats" separately from "hard-boiled detective novels." Once this tree is built, two separate features benefit: recommendations and search help. For recommendations, it tracks which branches of the tree you interact with, so it can suggest products that fit your taste, not just what's most popular.
For search, instead of just matching your exact words, the AI understands your intent. Type a vague query, and it suggests smarter refinements, like narrowing from "sneakers" to "running shoes under $100." It also guides you toward products the store actually sells, reducing the frustrating loop of clicking a suggestion that leads nowhere. The researchers tested this on a real shopping platform and saw measurable improvements: shoppers added more items to their carts, smaller brands got noticed more, and people reached a product they could buy with fewer detours.
This isn't a feature you can use today, but it shows a future where online catalogs work like a helpful in-store assistant. One honest limitation: the system needs a lot of high-quality product data to build that family tree, so smaller stores might struggle to see the same benefits. Still, it points toward a simpler, faster, and more surprising online shopping experience.
- A single AI-generated "product map" helps stores understand that different listings can mean the same thing, making searches and recommendations smarter.
- In tests, shoppers added more items to their carts and spent less time searching, while smaller brands gained more visibility.
- The approach works for both recommendation feeds and search suggestions, reducing the usual disconnect between the two.
Why It Matters
Better product suggestions and search mean less scrolling, faster purchases, and a fairer chance for smaller brands to be discovered.