GPASAN
The code and dataset for our paper “Modeling Periodic Pattern with Self-Attention Network for Sequential Recommendation”
Install / Use
/learn @hungrypig/GPASANREADME
GPASAN
We setup our experiment on an Nvidia Titan GPU and 256G memory on CentOs.
Experiment Environment
- python 3.6.5
- tensorflow-gpu 1.12
- numpy
Project Struct
GPASAN/data/
-
taobao_buy_8_time_4.zip ------ the dataset of taobao buy, Each line is in format: UserID \t ItemID \t CategoryId \t Time. we don't use the category information, It's just for catering to some other datasets' format. Unzip the taobao_buy_8_time_4.zip in this directory before running the program.
-
adj_matrix/taobao_buy_8/
- s_norm_adj_mat.npz ------ the preprocessed adjacent matrix of the item-item graph
- taobao_buy_8.txt ------ a temporary file
GPASAN/
- main.py ------ this is the program entry
- model_time_two_channel_gcn.py ------ the major PGASAN model
- modules_time.py ------ the periodic-attenuated self-attention network module
- sampler_time_repeat.py ------ The file used to create the training data
- util_periodic_two_channel.py ------ some helper and test functions we used
Recommended Setup
You can run the main.py directly for easily running the program.
If you run the code on linux, just running the following command:<br>
<br>
python main.py
