Skip to content

Recommendation Systems

Goal: Identify item(s) most relevant to a user

Types

Focus Free from New User Cold Start Problem Free from New Item Cold Start Problem Personalized Contextual
(eg: time of day)		 Scalable Rank
Popularity-based Top items ✅ ❌ ❌ ✅ ✅ ✅
Classifier Predict if each item would be liked or not ✅ ✅ ✅ ❌ ❌
Content-based filtering Recommend similar items ❌ ❌ ✅ ✅ ⚠️ ✅
User-based Recommend items that similar users liked
Collaborative Item-based filtering Recommend similar items, given co-occurence of purchases ❌ ❌ ✅ ✅ ⚠️ ✅
Collaborative User-based filtering Recommend items that similar users liked, given purchase history ❌ ❌ ✅ ✅ ⚠️ ✅
Matrix Factorization/
Model-based collaborative filtering		 Perform collaborative filtering on a low-dimensional space

Split original sparse matrix into constituent dense
- learnt embedding space for user matrix
- learnt embedding space for item matrix		 ❌ ❌ ✅ ✅
Last Updated: 2024-12-26 ; Contributors: AhmedThahir, web-flow

Comments