Skip to content

pykeen/ranking-metrics-manuscript

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

62 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Ranking Metrics Manuscript Supplement

arXiv Zenodo DOI

This repository contains analysis and supplementary information for A Unified Framework for Rank-based Evaluation Metrics for Link Prediction, non-archivally submitted to GLB 2022.

📣 Main Results 📣 There's a dataset size-correlation for common rank-based evaluation metrics like mean rank (MR), mean reciprocal rank (MRR), and hits at k (H@K) that makes them difficult to compare across datasets. We used the expectation, maximum, and variance of each metric to define adjusted metrics that don't have a dataset size-correlation and are more easily comparable across datasets.

Results

🖼️ Figure Summary 🖼️ While the MRR on, e.g., Nations and WN18-RR appears similar for ComplEx, the adjusted index reveals that when adjusting for chance, the performance on (the larger) WN18-RR is more remarkable. The z-adjusted metric allows an easier direct comparison against the baseline that suggests the results on smaller datasets are less considerable, despite achieving better unnormalized performance.

♻️ Reproduction

After installing tox with pip install tox, do the following:

  1. tox -e collate to build the combine results files
  2. tox -e plot to summarize the results files as plots

👋 Attribution

📖 Citation

@article{hoyt2022metrics,
    archivePrefix = {arXiv},
    arxivId = {2203.07544},
    author = {Hoyt, Charles Tapley and Berrendorf, Max and Gaklin, Mikhail and Tresp, Volker and Gyori, Benjamin M.},
    eprint = {2203.07544},
    month = {mar},
    title = {{A Unified Framework for Rank-based Evaluation Metrics for Link Prediction in Knowledge Graphs}},
    url = {http://arxiv.org/abs/2203.07544},
    year = {2022}
}

⚖️ License

The code in this package is licensed under the MIT License. The model, data, and results are licensed under the CC Zero license.

🎁 Support

This project has been supported by several organizations (in alphabetical order):

🏦 Funding

This project has been funded by the following grants:

Funding Body Program Grant
DARPA Young Faculty Award (PI: Benjamin Gyori) W911NF2010255
German Federal Ministry of Education and Research (BMBF) Munich Center for Machine Learning (MCML) 01IS18036A
Samsung Samsung AI Grant -