Efficient User Sequence Learning for Online Services via Compressed Graph Neural Networks

Instructions

1. Download the datasets in data folder and unzip it.

2. Install dependencies:

   pip install -r requirements.txt

3. Run ECSeq on Bike dataset:

   python ECSeq_traffic.py --dataset=bike_nyc --method=ECSeq

   The results will be printed, and also saved in results.txt; models will be saved in model folder; figures of evaluation results will be saved in fig folder.

   If you want to retrain the models, please empty the model folder first.

Arguments Description

Name	Default value	Description
dataset	bike_nyc	Dataset file name, can be chosen from {'bike_nyc', 'pems_bay'}.
seq_backbone	lstm	Sequence embedding extractor backbone, can be chosen from {'lstm', 'transformer'}.
gnn_backbone	GraphSAGE	Graph mining backbone, can be chosen from {'GraphSAGE', 'GraphSAGE_max', 'GCN', 'GAT'}.
method	ECSeq	'ECSeq': use ECSeq framework; 'batchGNN': don't use ECSeq and train GNN on graph batchs.
compress	kmeans_no	Graph compression algorithm, can be chosen from {'kmeans', 'AGC', 'Grain', 'RSA'}.
n_cluster	100	Number of clusters/new nodes.

References

• pytorch_geometric: https://github.com/pyg-team/pytorch_geometric
• UCTB: https://github.com/uctb/UCTB
• R-transformer: https://github.com/DSE-MSU/R-transformer
• graph coarsening: https://github.com/loukasa/graph-coarsening
• AGC: https://github.com/karenlatong/AGC-master
• Grain: https://github.com/zwt233/Grain