Stanford Question-Answering Dataset (v2.0) - parsed completely in ms-excel file.

The SQuAD 2.0 dataset is available online in JSON format Link There are several csv files for SQuAD 2.0 available for download on Kaggle etc. but none of them have all the attributes related to a question (as given in json files). Also, most of the datasets tend to ignore the unanswerable questions of SQuAD 2.0. The parsing of JSON files can be a task for beginners, so I have taken the opportunity of uploading the completely parsed SQuAD 2.0 dataset in excel files (both training and development sets).

The Training data has following columns:

1. Title
2. Context
3. Question
4. Id
5. Answer
6. Answer start
7. Plausible Answer
8. Plausible Answer Start
9. Is_impossible

The Development set has following columns:

1. Title
2. Context
3. Question
4. Id
5. Answer
6. Answer start
7. Plausible Answer
8. Plausible Answer Start
9. Is_impossible

The SQuAD 2.0 dataset has about 50,000 questions that are unanswerable and for such questions, the training and development dataset has plausible answer and plausible answer start options.

Train_set details: Total: 130319 unique questions Unanswerable questions: 43498 Answerable: 86821

Dev_set details: Total: 11873 unique questions Unanswerable questions: 5945 Answerable questions: 5928 15 unanswerable questions have no plausible answers given in the dataset.

Feel free to use the dataset for R&D purposes! Thank you!

Important Note: This dataset is the original property of Rajpurkar et al. (2018) and I haven't made any new changes in it so don't forget to cite Rajpurkar et al. (2018) when using this dataset.

Cite: Rajpurkar, P., Jia, R., & Liang, P. (2018). Know what you don't know: Unanswerable questions for SQuAD. arXiv preprint arXiv:1806.03822.