A type of recommendation system that tries to recommend top N products that a user will like. There are different approaches to implementation.
Item-based collaborative filtering
Top level view
Item-based collaborative filtering.canvas
The process looks like this:
- Individual interests: a user shows interest in some product.
- Item similarities: we collect information about similar items. Products are considered similar when users often by them together.
- Candidate generation: take products user interested in and find similar products
- Candidate ranking: rank products based on how often they’re showing up, how popular they are, etc.
- Filtering: filter out products that user already purchased and take top N
- Show Top-N recommendations to the user. Simply put: if user A likes product A and user B likes product A and product B then user A might also like product B.
This process is called item-based collaborative filtering item-based - it’s based on item similarities collaborative - we take into account other people actions to recommend items to the user