0Gollum / lessr

Handling Information Loss of Graph Neural Networks for Session-based Recommendation

LESSR

A PyTorch implementation of LESSR (Lossless Edge-order preserving aggregation and Shortcut graph attention for Session-based Recommendation) from the paper:
Handling Information Loss of Graph Neural Networks for Session-based Recommendation, Tianwen Chen and Raymong Chi-Wing Wong, KDD '20

Requirements

• PyTorch 1.6.0
• NumPy 1.19.1
• Pandas 1.1.3
• DGL 0.5.2

Usage

1. Install the requirements.
   If you use Anaconda, you can create a conda environment with the required packages using the following command.

   conda env create -f packages.yml

   Activate the created conda environment.

   conda activate lessr
   

2. Download and extract the datasets.

   • Diginetica
   • Gowalla
   • Last.fm

3. Preprocess the datasets using preprocess.py.
   For example, to preprocess the Diginetica dataset, extract the file train-item-views.csv to the folder datasets/ and run the following command:

   python preprocess.py -d diginetica -f datasets/train-item-views.csv

   The preprocessed dataset is stored in the folder datasets/diginetica.
   You can see the detailed usage of preprocess.py by running the following command:

   python preprocess.py -h

4. Train the model using main.py.
   If no arguments are passed to main.py, it will train a model using a sample dataset with default hyperparameters.

   python main.py

   The commands to train LESSR with suggested hyperparameters on different datasets are as follows:

   python main.py --dataset-dir datasets/diginetica --embedding-dim 32 --num-layers 4
   python main.py --dataset-dir datasets/gowalla --embedding-dim 64 --num-layers 4
   python main.py --dataset-dir datasets/lastfm --embedding-dim 128 --num-layers 4

   You can see the detailed usage of main.py by running the following command:

   python main.py -h

5. Use your own dataset.

   1. Create a subfolder in the datasets/ folder.
   2. The subfolder should contain the following 3 files.
      • num_items.txt: This file contains a single integer which is the number of items in the dataset.
      • train.txt: This file contains all the training sessions.
      • test.txt: This file contains all the test sessions.
   3. Each line of train.txt and test.txt represents a session, which is a list of item IDs separated by commas. Note the item IDs must be in the range of [0, num_items).
   4. See the folder datasets/sample for an example of a dataset.

Citation

If you use our code in your research, please cite our paper:

@inproceedings{chen2020lessr,
    title="Handling Information Loss of Graph Neural Networks for Session-based Recommendation",
    author="Tianwen {Chen} and Raymond Chi-Wing {Wong}",
    booktitle="Proceedings of the 26th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD '20)",
    pages="1172-–1180",
    year="2020"
}