ArxivQA: Multimodal Question Answering on Scientific Figures from ArXiv Papers

MMInstruction created a dataset for multimodal question answering, likely pairing images of scientific figures from arXiv papers with multiple-choice questions. The dataset was last updated on March 5, 2024. Each example includes an image path and a set of answer options, suggesting a focus on visual reasoning in academic contexts.

Use Cases

• Training multimodal LLMs to answer questions based on scientific figures mentioned in the description.
• Benchmarking visual reasoning capabilities of AI models on academic content from arXiv.
• Fine-tuning models for instruction-following tasks that require interpreting charts and diagrams.
• Researching the intersection of natural language processing and computer vision for scholarly documents.

Strengths

• Dataset is explicitly designed for multimodal question answering, combining images and text.
• Examples are structured with unique IDs, image paths, and multiple-choice options.
• Last update timestamp (2024-03-05) is provided, indicating recent maintenance.

Limitations

• Row count, file size, and column definitions are unknown, limiting suitability assessment.
• Column-level documentation is absent; field semantics must be inferred after download.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

MMInstruction on Hugging Face.

Collection Method

Likely constructed from arXiv paper figures and corresponding textual questions/answers.

Freshness

Last updated 2024-03-05 05:58:49.