- Published in
- Computation + Journalism Symposium, Columbia University (C+J 2015)
- Paper
- Research
- Northwestern University Intelligent Information Laboratory
Understanding a news story usually means knowing things the article does not supply: its background, who the people and organizations in it are, how its events compare with earlier ones, how it looks from other sides. The paper argues that much of that context can be supplied automatically, and describes the architecture the InfoLab abstracted from a decade of such systems, among them Tell Me More and LocalSavvy.
The architecture has six phases: analyze the content, form queries from it, choose which sources to ask, run the queries and collect the results, rank and extract from them, and present them beside the original. A Python toolkit covers each phase, so a new application is a matter of choosing components rather than starting over.
Three browser extensions were built on it. TweetTalk finds personal reactions to a story on Twitter and filters out the news organizations. Stakeholder Tweetback finds the verified accounts of the people and organizations in a story and shows what they have posted about it. Readdit finds the Reddit threads discussing the story. The paper closes with the next step, a platform that non-programmers could configure to define the context they want, and the editorial transparency that would bring.
The same work was presented at IUI 2015, and the toolkit was released as NUinfolab/context.