MovieLens1M 基于深度学习的电影推荐系统
使用MovieLens1M数据集(data can be downloaded from https://grouplens.org/datasets/movielens/) ,实现了Auto Encoder (AE), Variational Auto Encoder (VAE), BERT提取电影名特征3种方法,对评分矩阵进行填补,继而对用户做出推荐。
代码建议在Google Colab环境下运行,代码中的目录请根据自己的实际目录进行修改。
本代码主目录和子目录如下:
/content/drive/Movie_lens/
--------- ml-1m (包含数据集的文件夹)
--------- auto encoder.ipynb
--------- BERT-based-recommender.ipynb
1 Models:
1.1 Auto Encoder
1.2 Variational Auto Encoder
1.3 BERT-based
2 Experimental Results:
2.1 MSE of training loss and validation loss of Auto Encoder
2.2 MSE of training loss and validation loss of Variational Auto Encoder
2.3 MSE of training loss and test loss of BERT-based
2.4 test MSE loss of different models
Model | test MSE loss |
---|---|
Auto Encoder | 1.0837 |
Variational Auto Encoder | 0.9956 |
BERT-based | 0.7507 |
LICENSE
Please refer to MIT License Copyright (c) 2020 YJiangcm