DAWQAS: A Dataset for Arabic Why Question Answering System

摘要

A why question answering system is a tool designed to answerwhy-questions posed in natural language. Several papers have been published on the problem of answeringwhy-questions. In particular, attempts have been made to analyze Arabic text and predict which passages are best candidates for thewhy-questions; employing different datasets with limited size and not publicly available. To overcome these limitations, this paper introduces the new publicly available dataset, DAWQAS: Dataset for ArabicWhyQuestion Answering System. It consists of 3205 ofwhyquestion-answer pairs that were first scraped from public Arabic websites, then texts were preprocessed and converted to feature vectors. Afterwards,why-answers were re-categorized based on their domains. Finally, the rhetorical relations’ probabilities based on discourse markers were computed for each sentence in the dataset. DAWQAS is a valuable resource for research and evaluation in language understanding. The new dataset is freely available athttps://github.com/masun/DAWQAS.