Scientific Literature · UniverseTBD
arxiv-abstracts-large
A corpus of 1.7 million arXiv paper abstracts for training scientific and astronomy language models.
What arxiv-abstracts-large is
arxiv-abstracts-large is a scientific text dataset compiled by UniverseTBD. It comprises 1.7 million scholarly article records from the arXiv preprint repository, spanning disciplines including physics, astronomy, computer science, statistics, electrical engineering, quantitative biology, and economics.
The dataset is structured to support the pretraining and fine-tuning of scientific language models. By gathering abstracts, titles, and associated metadata across multiple quantitative domains, it enables computational researchers to build natural language processing models and analyse cross-disciplinary connections.
How it was made
The dataset stores scholarly literature records in JSON format with text modality. Individual entries include metadata and text features such as article titles, authors, subject categories, abstracts, and references to full-text PDFs.
Structured for text generation pipelines, the records are formatted to interface directly with common data handling and machine learning frameworks, including datasets, pandas, Polars, and mlcroissant.
Key facts
- Publisher
- UniverseTBD
- Dataset size
- 1.7 million records
- Data format
- JSON
- Language
- English
- Supported task
- Text generation
- Subject areas
- Physics, astronomy, computer science, statistics, electrical engineering, quantitative biology, economics
What you can use it for
- Language model pretraining
- Pretraining foundational language models on a broad corpus of physical and computational science abstracts.
- Domain adaptation for astronomy
- Fine-tuning models on domain-specific scientific text for downstream natural language tasks in astronomy and astrophysics.
- Cross-disciplinary literature exploration
- Analysing interconnections and thematic overlaps across multiple quantitative fields using abstracts and subject categories.
Explore the data
The schema, real rows, full-text search and a SQL filter — read in your browser, nothing downloaded.
Explore
Read the real rows without downloading anything
Reading rows…
LiveRead from UniverseTBD/arxiv-abstracts-large at the moment you asked. Nothing is cached or stored — every row above came from that request.
How the columns are distributed
Over 2,292,057 rows of default/train
abstract
string_text- min
- 6
- median
- 897
- mean
- 937.27
- max
- 6,091
authors
string_text- min
- 3
- median
- 44
- mean
- 71.69
- max
- 36,662
authors_parsed
list- min
- 1
- median
- 3
- mean
- 4.40
- max
- 2,832
categories
string_text- min
- 5
- median
- 13
- mean
- 14.96
- max
- 125
comments
string_text- min
- 1
- median
- 41
- mean
- 60.90
- max
- 1,150
23.6% empty
doi
string_text- min
- 7
- median
- 26
- mean
- 25.16
- max
- 153
50.7% empty
id
string_text- min
- 9
- median
- 10
- mean
- 10.58
- max
- 16
journal-ref
string_text- min
- 1
- median
- 32
- mean
- 40.28
- max
- 557
64.7% empty
license
string_label- http://arxiv.org/licenses/nonexclusive-distrib/1.0/67%
- http://creativecommons.org/licenses/by/4.0/9%
- http://creativecommons.org/licenses/by-nc-nd/4.0/1%
19.8% empty
report-no
string_text- min
- 1
- median
- 15
- mean
- 19.64
- max
- 479
92.3% empty
submitter
string_text- min
- 1
- median
- 14
- mean
- 14.44
- max
- 64
0.66% empty
title
string_text- min
- 1
- median
- 70
- mean
- 73.87
- max
- 382
versions
list- min
- 1
- median
- 1
- mean
- 1.58
- max
- 187
Strengths and limits
Good at
- Provides 1.7 million scholarly abstracts covering diverse physical, computational, and statistical disciplines.
- Specifically structured for pretraining and fine-tuning language models on scientific literature.
- Contains key article metadata alongside abstracts, including titles, author lists, and category tags.
Watch out for
- The corpus is limited to English-language scientific preprints.
- Coverage is focused on arXiv subjects, leaving broader life and biomedical sciences largely unrepresented outside quantitative biology.
Questions
Can I look inside arxiv-abstracts-large without downloading it?
Yes. This page reads its schema, real rows and per-column statistics in the browser, with full-text search and a SQL filter. Nothing is downloaded to your machine.
What licence is arxiv-abstracts-large under?
Its authors declare afl-3.0. Read the terms on the original page before reusing the data.
Which scientific fields are represented in the dataset?
The records are drawn from arXiv, encompassing physics, astronomy, computer science, statistics, electrical engineering, quantitative biology, and economics.
What information does each record contain?
Each record provides core article metadata including the title, author listing, subject categories, abstract text, and references to full-text PDFs.
What machine learning applications is this dataset intended for?
It is formatted for text generation tasks, notably the pretraining and fine-tuning of scientific and astronomy language models.
Which data processing libraries support the dataset format?
The dataset is provided in JSON format and is compatible with Hugging Face datasets, pandas, Polars, and mlcroissant.
Written from arxiv-abstracts-large's own dataset card. Statements citing a figure that does not appear there are removed before publication. Licence and running details come from our registry.