Standard Collaborative Filtering vs Hybrid Light FM

Question

I'm new to recommender system and trying to understand the fundamental difference between standard collaborative filtering (CF) and hybrid methods like LightFM. As I researched online, most of the posts mentioned hybrid method combines both CF and content-based method. But from a matrix/math standpoint, LightFM also learns item-user interaction embedding, like CF. How are they different?

Thank you so much in advance.

Answer 1

Based on Kula's [paper](https://arxiv.org/pdf/1507.08439.pdf, LightFM starts from a CF matrix factorization algorithm while also learning both users and items embeddings in the process (if such data available). However, if no user/item features are provided to the model, its behaviour will be that of a MF (Matrix Factorization) method.