Linking Wikipedia Events to Past News

We consider the task of linking Wikipedia events to relevant news articles from the past. Descriptions of events are abundant in Wikipedia and systematically curated in year, decade, and century articles. To address this task, we develop a two-stage cascade approach that builds a query model from temporal expressions in a set of initially retrieved documents. As baselines we consider several methods that integrate publication dates and/or temporal expressions into a language modeling approach. Our experimental evaluation on 50 randomly sampled Wikipedia events with crowd-sourced relevance assessments shows that the two-stage cascade approach outperforms the baselines. Our experimental testbed of queries and relevance assessments is made publicly available.