|
--- |
|
license: apache-2.0 |
|
--- |
|
|
|
Task: MCQ with multiple correct answers. |
|
Dataset: Recommendation of datasets to validate a research question. |
|
|
|
This dataset is derived from the [DataFinder](https://aclanthology.org/2023.acl-long.573/) dataset. We curate the abstracts of each dataset from [PapersWithCode](https://paperswithcode.com/datasets). |
|
Given is a short `query` discussing a research question, and keyphrases relevant the query. |
|
The original training set of the DataFinder dataset has positive and negative candidates for each query, to train a contrastive model. |
|
We objective is to convert the dataset into a MCQ question-answering task with multiple correct answers. We also add the abstracts from the research papers introducing the datasets so that context can be provided to the models. |
|
|
|
|
|
To reproduce the construction of this dataset, please visit [https://github.com/shruti-singh/scidata_recommendation](https://github.com/shruti-singh/scidata_recommendation). |
|
|
|
Please note that the query instances in this dataset have no intersection with the [`dataset_recommendation_mcq_sc`](https://huggingface.co/datasets/shrutisingh/dataset_recommendation_mcq_sc) dataset. [`dataset_recommendation_mcq_sc`](https://huggingface.co/datasets/shrutisingh/dataset_recommendation_mcq_sc) is a variant of this MCQ question-answering task with only single correct answer. |
