license: cc-by-4.0
task_categories:
- text-classification
- zero-shot-classification
tags:
- relation-classification
- relation-extraction
- few-shot
- domain-adaptation
- business
- finance
language:
- en
size_categories:
- 1K<n<10K
Dataset Card for CORE: A Few-Shot Company Relation Classification Dataset for Robust Domain Adaptation.
CORE includes 4,708 instances of 12 relation types with corresponding textual evidence extracted from company Wikipedia pages. It contains an annotated NOTA (none-of-the-above) category.
Dataset Details
Dataset Description
We introduce CORE, a dataset for few-shot relation classification (RC) focused on company relations and business entities. CORE includes 4,708 instances of 12 relation types with corresponding textual evidence extracted from company Wikipedia pages. Company names and business entities pose a challenge for few-shot RC models due to the rich and diverse information associated with them. For example, a company name may represent the legal entity, products, people, or business divisions depending on the context. Therefore, deriving the relation type between entities is highly dependent on textual context. To evaluate the performance of state-of-the-art RC models on the CORE dataset, we conduct experiments in the few-shot domain adaptation setting. Our results reveal substantial performance gaps, confirming that models trained on different domains struggle to adapt to CORE. Interestingly, we find that models trained on CORE showcase improved out-of-domain performance, which highlights the importance of high-quality data for robust domain adaptation. Specifically, the information richness embedded in business entities allows models to focus on contextual nuances, reducing their reliance on superficial clues such as relation-specific verbs. In addition to the dataset, we provide relevant code snippets to facilitate reproducibility and encourage further research in the field.
Dataset Sources [optional]
- Repository: https://github.com/pnborchert/CORE
- Paper: https://arxiv.org/abs/2310.12024
Dataset Structure
The dataset is split in training and test instances with overlapping relation types. Relation types inlcuded in the test set should be excluded from the training set in the episode sampling procedure sample_configuration.py.
train: Contains 4000 training instances and 12 relation types.test: Contains 708 instances and 12 relation types.relation_description: Textual descriptions of the relation types.
Citation
@misc{borchert2023core,
title={CORE: A Few-Shot Company Relation Classification Dataset for Robust Domain Adaptation},
author={Philipp Borchert and Jochen De Weerdt and Kristof Coussement and Arno De Caigny and Marie-Francine Moens},
year={2023},
eprint={2310.12024},
archivePrefix={arXiv},
primaryClass={cs.CL}
}