license: mit
AstroLLMs Gold Benchmark Dataset
This dataset is a collection of queries that astronomers asked to an astronomy research Slack chatbot. Along with the questions, there are open coding labels determined by a team of researchers and expert astronomer answers to these queries. Astronomers were asked to respond using citations and without the help of Large Language Models. This dataset of answers and responses is called the "Gold Benchmark Dataset".
Dataset Structure
The columns of this dataset include Query, Open Coding (the label that was assigned about the type of query asked), Answer (by expert astronomers), and First through Fifth Relevant Paper (the research paper related to the expert answer).
Citation
Please refer to the paper below and cite if this dataset was used in your work.
Hyk, A., McCormick, K., Zhong, M., Ciucă, I., Sharma, S., Wu, J. F., Peek, J. E. G., Iyer, K. G., Xiao, Z., & Field, A. "From Queries to Criteria: Understanding How Astronomers Evaluate LLMs," Conference on Language Modeling, 2025.