Extents based Headline

headline() produces text fragments with query terms to help users identify search results. It's tempting to use extents, which we already use to compute ranks.

  • Get extents
  • Order extents by importance
  • Choose 2-3 extents

We want headline to be informative, so extents should be dense (more query terms in shorter text) and dispersed (more different query terms). Heuristics: