This is the Arguana dataset from the BERDS benchmark. (Paper link: )The purpose of this dataset is evaluating diversity of retrieval systems given subjective questions.
Each instance consists of a question and a list of valid perspectives (opinions) for the question."Type" indicates the number of perspectives a question has. "Binary" questions come with two perspectives, while "Multi" questions come with more than two.All the instances in this dataset is "binary".
We repurpose the Arguana… See the full description on the dataset page:
https://huggingface.co/datasets/timchen0618/Arguana.
