BoxTaxo

This is the implementation of our paper "A Single Vector Is Not Enough: Taxonomy Expansion via Box Embeddings", published in WWW'23.

Data

The original data used could be access from SemEval-2016 Task 13: Taxonomy Extraction Evaluation.

We also provide our processed data under the data folder. Credits to repo!

Requirement

All required packages could be found in requirement.txt (generated by pip freeze).

How To Run?

cd ./src

python main.py

Key Arguments Interpretation:

• --dataset: Dataset option: environment or science.
• --embed_size: Dimension of box embeddings.
• --margin : Margin for containing loss ($\delta$ in paper).
• --epsilon: Margin for negative contain ($\epsilon$ in paper).
• --size: Minimum box ($\phi$ in paper).

The results (log, predictions, learned boxes) are stored under the result folder.

Contact

Song Jiang [email protected]

Bibtex

@inproceedings{boxtaxo,
  title={A Single Vector Is Not Enough: Taxonomy Expansion via  Box Embeddings},
  author={Song Jiang, Qiyue Yao, Qifan Wang, Yizhou Sun},
  booktitle={Proceedings of The Web Conference},
  year={2023}
}