license: cc-by-4.0
task_categories:
- text-generation
- text-classification
- zero-shot-classification
- sentence-similarity
language:
- en
size_categories:
- 10K<n<100K
ArXiv-10

"ArXiv-10" dataset consists of titles and abstracts extracted from 100k scientific papers on ArXiv, covering ten distinct research categories. These categories span subfields of computer science, physics, and mathematics. To ensure consistency and manageability, the dataset is downsampled to precisely 10k samples per category. This dataset provides a practical resource for researchers and practitioners interested in LLM. What is different about this dataset is the high complexity and domain-specific terminology, which currently poses significant challenges for LLM models. This specialized vocabulary found in research papers requires LLMs or LMs to have a deep understanding of context and semantic relationships.
Please cite the source when using this dataset:
@inproceedings{farhangiprotformer,
title={Protoformer: Embedding Prototypes for Transformers},
author={Farhangi, Ashkan and Sui, Ning and Hua, Nan and Bai, Haiyan and Huang, Arthur and Guo, Zhishan},
booktitle={Advances in Knowledge Discovery and Data Mining: 26th Pacific-Asia Conference, PAKDD 2022},
year={2022},
organization={PAKDD}
}