Get AUC 0.794 at Movielens 20M dataset
python3
tensorflow 1.14 or 1.15
step0, download dataset from https://download.csdn.net/download/guotong1988/85505311
step1, run data/preprocess.py
step2, run train.py
| Method | AUC |
|---|---|
| Deep Interest Network | 0.73 |
| Wide & Deep | 0.73 |
| Our MLP | 0.792 |
| Our transformer | 0.794 |
Our method focus on the data preprocess step for the Movielens dataset. In detail, we get the top-15 tags for each user and top-15 tags for each item/movie. We view each tag as an id.