Recommender Systems Explained
Understand how recommendations are made, evaluated and kept from narrowing what people see
A taste of a lesson
Our new recommender beat the old one offline on precision at 10, but the A/B test showed no change. How?
This happens often. Your offline data was logged under the old system, so it mostly records reactions to items the old system chose to show. A new model that recommends different items gets little credit for them offline, while one that imitates the old system scores well. Precision at 10 also may not match what drives your online metric, such as long term engagement. Look at how different the new recommendations actually are, and consider metrics that account for exposure. What share of the new model's top 10 had users ever been shown?
Written by the teacher as an example. In your lesson the tutor answers your own questions, and like any AI it can be wrong.
What you will be able to do
- Compare content based and collaborative filtering approaches
- Explain matrix factorisation as learning user and item vectors
- Handle implicit feedback and cold start situations
- Evaluate recommenders offline with ranking metrics and know their limits
- Discuss popularity bias, feedback loops and diversity
Lesson plan
- 1 Two families of recommenders Compare content based and collaborative approaches. Start
- 2 Neighbourhood methods Recommend with item to item and user to user similarity. Start
- 3 Matrix factorisation Learn user and item vectors whose dot products predict preference. Start
- 4 Implicit feedback and cold start Work with clicks and views and handle new users and items. Start
- 5 Evaluating recommendations Measure ranking quality offline and online. Start
- 6 Feedback loops and responsibility Consider diversity, popularity bias and user agency. Start
Try asking
About this tutor
An intermediate tutor for data scientists, product people and engineers curious about how recommendation systems work. You will compare content based and collaborative approaches, build intuition for matrix factorisation as learning user and item vectors, and handle implicit feedback such as clicks and views. Lessons cover the cold start problem, popularity bias, two stage retrieval and ranking, offline metrics like precision at k and NDCG, and why online experiments are still needed. The last lesson looks at feedback loops, diversity and the responsibility that comes with shaping what people see.
Reviews
Students can review a tutor after a paid lesson. Nobody has yet.
About the teacher
Linear algebra for AI, with geometry first and notation second
9 tutors 338 lessons taught Sample
I teach the linear algebra behind modern AI: vectors, matrices, similarity, eigenvectors and the methods built on them, such as PCA, clustering and recommender systems. I trained in applied mathematics and later worked on search and recommendation features, so I like to connect each idea to something a real system does. My lessons begin with pictures and small numbers you...
See Katarzyna's profile and tutorsMore like this
Other tutors on the same or nearby topics.