One way we could integrate the data into scoring is by using search clickthrough's to label pages:

• Record the original query, the page title and the position of the clicked search result.
  • Google does this by sending all search results through a redirect
  • Yahoo seems to be opening links in a new tab, and sending a beacon in the SERP
  • I didn't see how duckduckgo is going about it, they might be sampling or something.
  • Bing looks to send a beacon to a previously clicked link on your next visit to bing
• Build up a dataset of the top terms that led to clicking through to an article. This will require some research to do right. Filtering will need to be correct to ensure we don't just add junk to the scores.
• Add a field to the documents in elasticsearch to hold these terms
• boost them at query time?

Caveats:

• This equates clicking a link to 'voting' for that links relevance for your search terms. That might be a valid assumption, but it probably doesn't hold for 100% of searches.
• probably plenty more