Personalisation and recommendation are one of the most most effective applications of machine learning.
They allow businesses to tailor products and services to individuals based on a combination of user behaviours and item features. Eugene Yan has written a useful review of some of the main approaches. He splits them into 5 groups:
➡ Embeddings + MLP: A good simple starting point
➡ Bandits: A way to balance exploration and exploitation
➡ Sequential: If you've got long user histories
➡ Graphs: Not much behaviour data but lots of item/user metadata
➡ User models: Generic embedding for multiple problems
🛎️ Why this matters: There have been a lot of developments in this field: here is a really useful map.