aitetic/arxiv-cs

arXiv Computer Science 2015–2024

This dataset contains arXiv records from Computer Science categories published between 2015 and 2024.

• Items: 1_432_025

The dataset consists of two JSONL files:

• arxiv-cs-2015-2024.jsonl

Both files are automatically combined into a single train split when the dataset is loaded through the Hugging Face datasets library.

File	Period
arxiv-cs-2015-2024.jsonl	2015–2024

Loading the dataset

Load the complete dataset:

from datasets import load_dataset

dataset = load_dataset("aitetic/arxiv-cs")

print(dataset)
print(dataset["train"][0])

Both JSONL files will be downloaded and combined into dataset["train"].

Loading the train split directly

from datasets import load_dataset

dataset = load_dataset(
    "aitetic/arxiv-cs",
    split="train",
)

print(dataset)
print(dataset[0])

Streaming

The dataset can be streamed without downloading and preparing the entire dataset locally:

from datasets import load_dataset

dataset = load_dataset(
    "aitetic/arxiv-cs",
    split="train",
    streaming=True,
)

for record in dataset.take(5):
    print(record)

Downloading the repository files

To download the repository in its original form, including both JSONL files and this README:

from huggingface_hub import snapshot_download

local_path = snapshot_download(
    repo_id="aitetic/arxiv-cs",
    repo_type="dataset",
)

print(local_path)

Intended uses

The dataset may be used for:

• causal language-model pretraining;
• continued pretraining on scientific text;
• scientific-text analysis;
• information extraction;
• semantic search and retrieval;
• document classification;
• topic modeling;
• analysis of Computer Science publications.

Users should independently evaluate whether the dataset is suitable for their intended application.

Data source

The records originate from arXiv and are associated with Computer Science subject categories.

This dataset repository is not affiliated with or endorsed by arXiv.

Limitations

The dataset may contain:

• incomplete or malformed records;
• duplicate or near-duplicate documents;
• withdrawn or revised papers;
• LaTeX markup;
• formatting inconsistencies;
• incomplete metadata;
• records from multiple Computer Science subcategories.

Additional cleaning and validation may be required before training or evaluation.

License

The repository uses license: other because individual arXiv papers may be distributed under different licenses selected by their authors.

Users should review the license associated with each paper, the arXiv terms of use, and any applicable copyright or redistribution requirements.

Citation

When using this dataset, please cite this repository and the original arXiv papers relevant to your work.

@dataset{aitetic_arxiv_cs_2015_2024,
  author       = {AItetic},
  title        = {arXiv Computer Science 2015--2024},
  year         = {2026},
  publisher    = {Hugging Face},
  url          = {https://huggingface.co/datasets/aitetic/arxiv-cs}
}