Multi-documents Non-factoid Question Answering Dataset.
This repository presents a dataset containing exclusively non-factoid questions for multi-document QA. The data set should contain for each question the relevant documents, the relevant information span in each document, and a correct answer.
For a question to be considered for the dataset, it should be non-factoid and complex enough to be multi-documents. Also, the question should be considered of the types definition, casual, or procedural. Each question contains at least two relevant documents that should be considered to the QA model to provide a relevant, complete and precise answer.
- Question
- Answer
- Multiple relevant documents.
- Relevant span of text for the answer.