Collaborative Filtering
Definition
Collaborative filtering recommends content based on similarities in the behavior of many users or objects.
The method uses interaction patterns such as ratings, purchases, or clicks without necessarily requiring detailed content features. Common variants are user-based, item-based, and matrix factorization.
Simply explained
Like saying: People with similar behavior also liked these things.
Why it matters in practice
Relevant for recommendations in shops, portals, and knowledge systems with sufficient interaction data.
Example from business
A B2B portal recommends documents based on similar usage patterns of other user groups.
Limits and pitfalls
- Cold start; popularity bias; feedback loops; treating implicit feedback as true preference.
Origin & history
Collaborative filtering developed with early recommender systems and was widely adopted by large platforms.
Use in mid-sized companies
- Possible application in SMEs: Supplement recommendations when sufficient interaction data and clear data protection rules are available.
Frequently asked questions
Does collaborative filtering need product features?
Not necessarily. What is cold start? New users or items have hardly any interaction data.
Sources
- Matrix Factorization Techniques for Recommender Systems · Koren et al. (2009), Matrix Factorization Techniques for Recommender Systems
- Bell & Volinsky (2009), zugängliche Originaltextkopie, Seite 31 · Koren, Bell & Volinsky (2009), zugängliche Originaltextkopie, Seite 31
Source check: 10./11.10.2026.
Editorially reviewed on: