Echoless-LP

Source code and dataset of the paper "Echoless Label-Based Pre-computation for Memory-Efficient Heterogeneous Graph Learning", which is accepted by AAAI 2026.

Homepage and Paper

• Homepage (Echoless-LP): https://github.com/CrawlScript/Echoless-LP
• Paper Access:
  • ArXiv: https://arxiv.org/abs/2511.11081

Dataset Download

For DBLP, IMDB, and Freebase (from the HGB benchmark), please refer to the official repository for download instructions:

• HGB benchmark: https://github.com/THUDM/HGB
• Unzip the downloaded files into the datasets directory.

For OGBN-MAG, the code will automatically download the dataset via the ogb package.

For OAG-Venue and OAG-L1-Field, we follow the dataset preparation instructions from the NARS baseline, with minor file renaming:

• Instructions: https://github.com/facebookresearch/NARS/tree/main/oag_dataset
• After generating the *.pk and *.npy files:
  • Place these files in the ./datasets/nars_academic_oag/ directory.
  • Rename graph_field.pk to graph_L1.pk.

Requirements

• Linux
• Python 3.7
• torch==1.12.1+cu113
• torchmetrics==0.11.4
• dgl==1.0.2+cu113
• ogb==1.3.5
• shortuuid==1.0.11
• pandas==1.3.5
• gensim==4.2.0
• numpy==1.21.6
• tqdm==4.64.1
• wandb==0.18.3

Run Echoless-LP

You can run Echoless-LP with the following command:

sh scripts/run_DBLP.sh

sh scripts/run_Freebase.sh

sh scripts/run_IMDB.sh

sh scripts/run_OGBN-MAG.sh

sh scripts/run_OAG-Venue.sh

sh scripts/run_OAG-L1-Field.sh

Cite

If you use Echoless-LP in a scientific publication, we would appreciate citations to the following paper:

@misc{hu2025echolesslabelbasedprecomputationmemoryefficient,
      title={Echoless Label-Based Pre-computation for Memory-Efficient Heterogeneous Graph Learning}, 
      author={Jun Hu and Shangheng Chen and Yufei He and Yuan Li and Bryan Hooi and Bingsheng He},
      year={2025},
      eprint={2511.11081},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/2511.11081}, 
}