What is collaborative filtering in fashion?
A recommendation method that suggests items based on what similar users chose, rather than on product attributes.
In short
Collaborative filtering is a recommendation method that suggests items based on what similar users chose, rather than on the product's own attributes. In fashion wholesale it can suggest styles to a retailer because comparable retailers ordered them together.
How does it work in practice?
The underlying idea is that people who agreed in the past will agree again. The system looks at patterns of orders, views or purchases across many users. If retailers who bought a certain knitwear style also tended to order a particular trouser, the system will suggest that trouser to a new buyer of the knit. It does not need to know what either product looks like.
There are two common approaches:
- User-based: find accounts similar to this one and suggest what they ordered.
- Item-based: find items frequently chosen together with the item in question.
Fashion uses include assortment suggestions in a B2B portal, 'frequently bought together' features in e-commerce and identifying cross-selling opportunities for sales agents.
Why does it matter?
Collaborative filtering captures real buying behaviour, including combinations that attribute rules would not predict. It can reveal how different store types build their assortments and help smaller accounts benefit from the choices of comparable retailers.
How does AI use it?
Collaborative filtering is a core technique in recommender systems and personalisation engines. Modern versions use machine learning to find hidden patterns, and are usually combined with content-based methods and clustering of accounts or customers.
Common pitfalls
- Cold start. Brand-new products that no one has bought yet cannot be recommended, which is a significant issue in seasonal fashion.
- Popularity bias. Bestsellers get recommended more, which can crowd out newness.
- Sparse data. Smaller brands or niche categories may have too few orders for reliable patterns.
- Context blindness. Results should be filtered for market, season and availability.
Frequently asked questions
What is the cold-start problem in fashion recommendations?
It is when a new product or new account has no history, so collaborative filtering cannot recommend it. Attribute-based methods and similar past styles are used to fill the gap.
How is collaborative filtering different from content-based filtering?
Collaborative filtering relies on the behaviour of similar users, while content-based filtering compares product attributes such as category, colour and fabric.
