[paper review] Sampling-Bias-Corrected Neural Modeling for Large Corpus Item Recommendations

1. Introduction

Candidate generation

2. Modeling Framework

Two-tower DNN

in-batch negatives (batch negative sampling)

3. Streaming Frequency Estimation

4. Neural Retrieval System for Youtube

Experiment

Reference

https://angrypark.github.io/recommender%20system/two-tower-model/
https://www.youtube.com/watch?v=FSDuo9ybv8s&t=1932s