LowFER

Code for the paper "LowFER: Low-rank Bilinear Pooling for Link Prediction", ICML 2020

Experiments

See experiments.txt for the commands to run the model with the best hyperparameters.

Requirements

The original codebase was implemented in Python 3.6.6. Required packages are:

numpy      1.15.1
pytorch    1.0.1

Scripts

Run the scripts from main directory as python -m scripts.filename:

toy_example.py: This script contains the toy dataset used to visualize Proposition 1 in the paper. It runs LowFER under the conditions specified and shows that it perfectly separates positive examples from negative examples.

bilinear_models_relations.py: This script runs the conditions presented in relation to bilinear models (section 4.3) with a toy setup and shows the equivalence between the LowFER version of other bilinear models and the true scoring functions of those models.

relation_results_analysis.py: Evaluates per relation metrics on WN18/RR dataset using LowFER as reported in the per relation analysis results (section 5.4). It requires a trained model, so please first run LowFER for WN18/RR as detailed in experiments.txt.

plots.py: Simple plots script used to generate the effect of k and de.

Update: A refactored code version can be found on the refactor branch with faster evaluation. The results are slightly less (in third decimal) than this implementation.

Citation

If you find our work useful, please consider citing:

@inproceedings{pmlr-v119-amin20a,
  title ={{L}ow{FER}: Low-rank Bilinear Pooling for Link Prediction},
  author = {Amin, Saadullah and Varanasi, Stalin and Dunfield, Katherine Ann and Neumann, G{\"u}nter},
  booktitle = {Proceedings of the 37th International Conference on Machine Learning},
  pages = {257--268},
  year = {2020},
  editor = {III, Hal Daumé and Singh, Aarti},
  volume = {119},
  series = {Proceedings of Machine Learning Research},
  month = {13--18 Jul},
  publisher = {PMLR},
  pdf = {http://proceedings.mlr.press/v119/amin20a/amin20a.pdf},
  url = {https://proceedings.mlr.press/v119/amin20a.html}
}

Also, check our follow-up work extending LowFER with time-aware and model-agnostic temporal representations for TKGC and the accompanying temporal knowledge graph embeddings framework ChronoKGE:

@inproceedings{dikeoulias-etal-2022-temporal,
    title = "Temporal Knowledge Graph Reasoning with Low-rank and Model-agnostic Representations",
    author = {Dikeoulias, Ioannis and Amin, Saadullah and Neumann, G{\"u}nter},
    booktitle = "Proceedings of the 7th Workshop on Representation Learning for NLP",
    month = may,
    year = "2022",
    address = "Dublin, Ireland",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2022.repl4nlp-1.12",
    doi = "10.18653/v1/2022.repl4nlp-1.12",
    pages = "111--120",
}

Acknowledgements

The code is based on the open-source code released by TuckER. If you find our work useful, please consider this link for the original code and cite their work.