A Non-official Implementation of "Deep Matrix Factorization Models for Recommender Systems"
See paper: http://www.ijcai.org/proceedings/2017/0447.pdf
If you use the codes for your paper as baseline implementation, please cite the link: https://github.com/hegongshan/deep_matrix_factorization
We use Keras with Tensorflow as the backend.
- Keras version: 2.3.0
- TensorFlow: 2.0.0
python dmf.py --dataset ml-1m --user_layers [512,64] --item_layers [1024,64] --epochs 100 --lr 0.0001
when epochs = 10 and lr = 0.001
| HR@10 | NDCG@10 | model file | |
|---|---|---|---|
| ml-1m | 0.5225 | 0.2930 | model/ml-1m_u[512, 64]_i[1024, 64]_256_1572343913.h5 |
Tips: Each epoch takes about an hour and a half.
If you are interested in DMF, you can try to set lr to 0.0001 and run 100 epochs.
And then, HR@10 and NDCG@10 should be closer to the results in this paper.
Last Update: November 10, 2020