Microsoft Research Sequential Question Answering is a dataset released by Microsoft Research in 2017 for sequential question answering tasks, designed to explore more natural conversational question-answering scenarios and address the overly complex and unnatural long questions in traditional semantic parsing.

The dataset contains 17,553 questions, divided into 6,066 sequences, with each question associated with cell positions in a table as answers. The data is based on the WikiTableQuestions dataset and was generated by crowdworkers decomposing the original questions.

Dataset Composition

The dataset is divided into the following three parts:

• Training set (train): 12,276 samples.
• Validation set (validation): 2,265 samples.
• Test set (test): 3,012 samples.

Each sample mainly contains the following fields:

• id: Unique identifier of the question sequence.
• annotator: Annotator identifier.
• position: Position of the question within the sequence.
• question: Question text.
• table_file: Associated table file.
• table_header: Table column names.
• table_data: Table data, stored as a two-dimensional array.
• answer_coordinates: Row and column coordinates of the cells containing the answers.
• answer_text: Answer text content.

Citation```bibtex

@inproceedings{iyyer-etal-2017-search, title = "Search-based Neural Structured Learning for Sequential Question Answering", author = "Iyyer, Mohit and Yih, Wen-tau and Chang, Ming-Wei", booktitle = "Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", month = jul, year = "2017", address = "Vancouver, Canada", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/P17-1167", doi = "10.18653/v1/P17-1167", pages = "1821--1831", }