An Efficient Subgraph-Inferring Framework for Large-Scale Heterogeneous Graphs

Authors: Wei Zhou, Hong Huang, Ruize Shi, Kehan Yin, Hai Jin

AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on five benchmark datasets demonstrate that Sub Infer effectively optimizes the training and inferring phase, delivering comparable performance to traditional HGNN models while significantly reducing time and memory overhead.
Researcher Affiliation Academia Wei Zhou, Hong Huang*, Ruize Shi, Kehan Yin, Hai Jin National Engineering Research Center for Big Data Technology and System Services Computing Technology and System Lab, Cluster and Grid Computing Lab School of Computer Science and Technology, Huazhong University of Science and Technology, Wuhan, 430074, China {weizhou2021, honghuang, rzshi, kehanyin, hjin}@hust.edu.cn
Pseudocode Yes (The complete allocation process is shown in Supplementary materials3 Algorithm 1).
Open Source Code Yes 3https://github.com/CGCL-codes/Sub Infer
Open Datasets Yes To validate the performance and efficiency of our framework, we train both our framework (Sub Infer) with the baseline on five benchmark datasets: Ogbn-mag (Wang et al. 2020a), DBLP (Yang et al. 2022), Pub Med (Yang et al. 2022), Yelp (Yang et al. 2022) and Freebase(Lv et al. 2021).
Dataset Splits Yes Finally, in the inferring stage, we collect the output results of all nodes to the master node for processing and verify the performance of the model on the validation set and test set.
Hardware Specification No The paper mentions 'distributed clusters' and 'distributed nodes' but does not specify any particular hardware models (e.g., GPU/CPU models, memory).
Software Dependencies No The paper mentions tools like 'Metis (Karypis and Kumar 1998)' and 'Louvain (Blondel et al. 2008)' but does not provide specific version numbers for any software dependencies.
Experiment Setup Yes The paper includes a 'Hyper-parameter Study' section that discusses and presents specific values for hyperparameters like 'ϱ' (number of subgraphs, e.g., p=100, p=200), 'σ' (ratio of global information, e.g., sigma=0.02), and 'α' (masking ratio, e.g., alpha=0.1). While it also mentions details in supplementary materials, these specific parameters and their values are discussed in the main text.