Bigger versus Similar: Selecting a Background Corpus for First Story Detection based on Distributional Similarity Archives · ADAPT: The Global Centre of Excellence for Digital Content and Media Innovation