FreeGEM (NeurIPS 2022)

This is the implementation for our NeurIPS 2022 paper:

Parameter-free Dynamic Graph Embedding for Link Prediction.
Jiahao Liu, Dongsheng Li, Hansu Gu, Tun Lu, Peng Zhang, Ning Gu.
The Thirty-sixth Conference on Neural Information Processing Systems (NeurIPS). 2022.

We provide the code which can be used to reproduce the results of Table 3: Accuracy comparison with state-of-the-art methods on two link prediction tasks.

Introduction

There are four folders in this project, namely data, preprocess, future-item-recommendation and next-interaction-prediction. We will introduce how to use them in turn.

Note that the model reads the processed dataset, so you must first use the script in preprocess folder to preprocess the raw datasets in data and obtain the processed dataset before running the model in future-item-recommendation and next-interaction-prediction.

We welcome you to contact the authors or open issues when you encounter any problems.

`data`

The raw datasets is saved in this folder, all datasets are public and available.

For future item recommendation task
- Video, Game: http://jmcauley.ucsd.edu/data/amazon/links.html
- ML-100K, ML-1M: https://grouplens.org/datasets/movielens/
For next interaction prediction task
- Wikipedia, Lastfm: http://snap.stanford.edu/jodie/

At the same time, this folder is also used to store the processed datasets. The processed datasets can be obtained through the script in preprocess folder.

`preprocess`

This folder contains 3 scripts for preprocessing datasets.

amazon.ipynb is used to preprocess Video and Game.
ml-100k.ipynb is used to preprocess ML-100K.
ml-1m.ipynb is used to process ML-1M.

`future-item-recommendation`

After running the script in preprocess to obtain the processed datasets:

To reproduce the results of FreeGEM *(with attr) on ML-100K in Table 3(a)

python main.py --dataset ml-100k --attr --beta 15 --dim0 1 --dim1 1 --alpha 3
To reproduce the results of FreeGEM *(with attr) on ML-1M in Table 3(a)

python main.py --dataset ml-1m --attr --beta 50 --dim0 4 --dim1 1 --alpha 3
To reproduce the results of FreeGEM *(no attr) on Video in Table 3(a)

python main.py --dataset video --beta 21 --dim0 128
To reproduce the results of FreeGEM *(no attr) on Game in Table 3(a)

python main.py --dataset game --beta 18 --dim0 256
To reproduce the results of FreeGEM *(no attr) on ML-100K in Table 3(a)

python main.py --dataset ml-100k --beta 60 --dim0 1
To reproduce the results of FreeGEM *(no attr) on ML-1M in Table 3(a)

python main.py --dataset ml-1m --beta 60 --dim0 8

`next-interaction-prediction`