SciGraphQA: Multi-Turn Question-Answering for Scientific Graphs

SciGraphQA contains synthetic multi-turn question-answering dialogues paired with scientific graphs, published by findalexli in 2024. It targets vision-language tasks by providing sequential reasoning data for graphical interpretation.

Use Cases

• Training Vision Transformers to extract data points from synthetic scientific graphs
• Fine-tuning LLMs on multi-turn dialogue sequences to improve visual context retention
• Evaluating VQA models on their ability to interpret graph axes and legends

Strengths

• Multi-turn dialogue structure for sequential reasoning
• Apache-2.0 open-source license
• Focus on scientific graph interpretation

Limitations

• Synthetic data generation may lack the noise and complexity found in organic scientific publications
• Lack of explicit record counts in the repository metadata

Provenance

Source

findalexli via GitHub

Collection Method

Synthetic generation

Freshness

Last updated October 2024.