Dataset of news articles for the Timeline Generation Problem. See (Holt et. al 2016).
The primary dataset is comprised of the set of crowd-annotated articles for our entities. We define gold-standard timelines to be comprised of the articles which are labeled 'valid' and 'very important'. Columns:
- entity: The name of the entity.
- URL: Article URL.
- valid: Crowd-annotated label. An article is 'valid' if it is concerned with a single event in the history of the entity.
- valid_conf: Confidence of above annotation.
- importance: Crowd-annotated label. One of 'not', 'somewhat' or 'very' important.
- importance_conf: Confidence of above annotation.
The secondary dataset is comprised of a number of entity-linked articles retrieved by querying Google News. Columns:
- entity: The name of the entity.
- index: The index of the article in the relevant Google News query.
- URL: Article URL.
- published: Publish date.