Collaborative filtering recommends content by finding patterns across users: if many viewers with tastes similar to yours watched and liked a title, it is likely a good recommendation for you, even without knowing anything about the content itself. It powered the classic wave of recommenders and remains a core technique.
Its strength is that it captures subtle, hard-to-describe taste patterns directly from behavior. Its main weakness is the cold-start problem: new users and new titles have little interaction data, so they are hard to place. Production systems combine collaborative filtering with content-based methods and metadata to cover the gaps and keep recommendations fresh.
The handbook · 7 volumes
Fora Soft Video Engineering Handbook
Collaborative filtering recommends what similar viewers watched, and it works only as well as the viewing signals you capture. Our seven volumes cover the video stack; for collaborative filtering start with Volume 6 on ML basics, model artifacts, and search over video, then Volume 7 on serving infrastructure and cost, with QoE signals in Volume 3. Written by Nikolay Sapunov, CEO at Fora Soft.
Improving collaborative filtering recommendations on your streaming service?
We will audit your viewing data and recommender setup and send changes with a plan to measure the lift. 250+ video products shipped since 2005.
