RepLab Summarization Dataset

Each topic consists of a different number of tweets posted by Twitter users. The collection comprises tweets about 31 entities from two domains: automotive and banking. As a result, our subset of RepLab 2013 comprises 71,303 English and Spanish tweets For each entity, tweets are groupped in topics and for each topic three different summaries are manually generated: abstractive english, abstractive spanish and extractive.

Please see the paper for further details. 2. Description of the contents of this package ./entities: This directory includes the information of each organization in order to create a summary. Each .xml file corresponds to an entity and includes the following information: -”Corpus entity”: Id of the entity. -”cluster”: each one of the topics of the entity. -"label": name of the topic. -"priority": level of relevance of the topic: Alert (the highest priority being a reputation alert, i.e., an issue that requires an immediate response from the entity), Midly_important (relevant for the entity, an intermediate priority) or unimportant (the lowest priority). -”tweet”: Information about the tweets. -"id": Id of the tweet. -"date": When the tweet was written. -"followers": Of the author of the tweet. -"polarity": Of the tweet. -"text": Text of the tweet. -"summary": Information about the summary: -"abstract_EN": Abstractive summary in English. -"abstract_ES": Abstractive summary in Spanish. -"tweet": Id of the tweet(s) selected for the extractive summary (if it is not filled, the extractive summary is the one of the tweets in the topic).