Towards Open Temporal Graph Neural Networks

Authors: Kaituo Feng, Changsheng Li, Xiaolu Zhang, JUN ZHOU

ICLR 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments on three real-world datasets of different domains demonstrate the superiority of our method, compared to the baselines.
Researcher Affiliation Collaboration Kaituo Feng Beijing Institute of Technology kaituofeng@gmail.com Changsheng Li Beijing Institute of Technology lcs@bit.edu.cn Xiaolu Zhang Ant Group yueyin.zxl@antfin.com Jun Zhou Ant Group jun.zhoujun@antfin.com
Pseudocode Yes Algorithm 2 OTGNet: Open Temporal Graph Neural Networks
Open Source Code No The paper does not provide any explicit statement or link indicating that the source code for the described methodology is open or publicly available.
Open Datasets Yes We construct three real-world datasets to evaluate our method: Reddit (Hamilton et al., 2017), Yelp (Sankar et al., 2020), Taobao (Du et al., 2019).
Dataset Splits Yes For each task, we use 80% nodes for training, 10% nodes for validation, 10% nodes for testing.
Hardware Specification Yes We perform our experiments using Ge Force RTX 3090 Ti GPU.
Software Dependencies No The paper mentions using the 'Adam optimizer' but does not specify version numbers for any software, libraries, or frameworks used (e.g., Python, PyTorch, TensorFlow).
Experiment Setup Yes For each task, we use 80% nodes for training, 10% nodes for validation, 10% nodes for testing. We use the Adam optimizer for training with learning rate η = 0.0001 on the Reddit dataset, learning rate η = 0.005 on the Yelp datasets and learning rate η = 0.001 on the Taobao datasets. For the Reddit dataset and the Yelp dataset, we train each task 500 epochs. For the Taobao dataset, we train each task 100 epochs. We set the dropout rate to 0.5 on all the datasets. The node classification head is a two-layer MLP with hidden size 128. The selected triad pairs per class M is set to 10 on all datasets. The sub-network extracting class-agnostic information is a two-layer MLP with hidden size 100.