ArXiv QA Dataset: 12,853 Question-Answer Pairs from Computer Science Paper Abstracts

12,853 question-answer pairs generated from arXiv paper abstracts, covering computer science research across multiple subfields. The dataset was created by Navyasri12355 for instruction-tuning language models on scientific literature understanding tasks. It was last updated on the Hugging Face platform on 2026-05-29.

Use Cases

• Instruction-tuning language models for scientific text understanding based on the described QA pairs.
• Benchmarking question-answering systems on computer science literature based on the structured questions and answers.
• Training models to generate concise answers from research paper abstracts based on the dataset's input-output structure.

Strengths

• Contains 12,853 QA pairs, providing a substantial corpus for training.
• Focuses on computer science research from arXiv, a major preprint repository.
• Designed specifically for instruction-tuning, a relevant task for modern LLMs.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is known, but the specific source papers, time range, and generation methodology details are not provided in the input.
• Last updated 2026-05-29 05:26:18; freshness should be verified.

Provenance

Source

arXiv paper abstracts

Collection Method

Synthetic generation; likely involves pairing structured questions with paper titles and abstracts.

Freshness

2026-05-29