Answer Reformulation Dataset Questions and Answers extracted from StackExchange. Answers are paired with their voice-friendly version. Please cite: Voice-based Reformulation of Community Answers. Simone Filice, Nachshon Cohen and David Carmel. The Web Conference 2020. Authors: - Simone Filice (filicesf@amazon.com) - Nachshon Cohen (nachshon@amazon.com) - David Carmel (dacarmel@amazon.com) Dataset Structure: The dataset is a csv file where each row corresponds to a sentence extracted from the best answer to a StackExchange question. The columns are: - url: the url of the answer - domain: the StackExchange domain (e.g., Economics, Gardening, Literature, etc.) - sentence: the text of the sentence - position: the sentence position within the answer - answer_html: the html of the entire answer - query: the short question (it was used to determine the voice-friendly version of the answer) - full_question: the entire question in StackExchange - full_question_html: the html of the entire question in StackExchange - question_asker_username: the StackExchange username of the question asker - answerer_username: the StackExchange username of the answer author - gain: the number of annotators voting this sentence as the most relevant sentence in the answer - is_best: whether this sentence was the most voted by the annotators as most relevant sentence in the answer. This is the label to be used for the single-sentence answer reformulation task - is_complete_answer_according_to_all_annotators: whether all the annotators considered this single sentence as a valid and complete answer for the query - is_complete_answer_according_to_who_voted_it: whether all the annotators who voted this sentence as the most relevant considered this single sentence as a valid and complete answer for the query - is_in_best_selection: whether the multi-sentence voice-friendly version of the answer selected by the annotators contains this sentence. This is the label to be used for the multi-sentence answer reformulation task - is_selected_perc: the percentage of annotators who included this sentence in their multi-sentence voice-friendly version of the answer.