The ScIRGen-Geo Dataset is a large-scale, task-oriented dataset designed for retrieval-augmented generation (RAG) in scientific research, with a focus on the geoscience domain. The corpus is fully bilingual (English ↔ Chinese), offering parallel content in both languages.
This dataset is introduced in the paper "ScIRGen: Synthesize Realistic and Large-Scale RAG Dataset for Scientific Research", which has been accepted by the KDD conference.… See the full description on the dataset page: https://huggingface.co/datasets/usail-hkust/ScIRGen-Geo.
