SelQa
SelQA is a benchmark for the question answering task based on the selection of the correct answer among several candidates. The dataset contains questions with context and candidate answers, often paraphrased, intended to assess the accuracy of QA models.
Description
The dataset SelQa includes over 400,000 examples in English, each example containing a question, a context (section or article), and a set of candidate answers from which the correct one must be selected. Data is provided in JSON and Parquet format.
What is this dataset for?
- Train question answering models based on the selection of answers
- Evaluate the ability of models to identify the correct response in a given context
- Improving QA systems for practical applications like virtual assistants
Can it be enriched or improved?
Yes, it is possible to add additional annotations on the types of questions, to integrate more varied paraphrases, or to extend the corpus to other languages to reinforce the robustness of the models.
🔎 In summary
🧠 Recommended for
- NLP researchers
- Virtual assistant developers
- AI students
🔧 Compatible tools
- Hugging Face Transformers
- PyTorch
- TensorFlow
- Jupyter notebooks
💡 Tip
Use the available paraphrases to train the robustness of the models in the face of linguistic variation.
Frequently Asked Questions
What is the size of the SelQA dataset?
The dataset includes approximately 404,776 examples, with a file size of approximately 68 MB.
Can SelQa be used for open questions or multiple choice questions only?
It is primarily designed for selecting the correct answer from multiple candidates, not for open-ended questions.
Is this dataset suitable for commercial use?
Yes, the Apache 2.0 license allows unrestricted commercial use.
