News for hackers. Stuff that matters.
Re:The Berkeley Document Summarizer: Learning-Based, Single-... by zuzun on Sunday October 02, 2016 @05:43AM (#12621642)
Summarizing news articles the way the bot does isn't really that difficult. You face the task of picking 5 sentences out of 20. Since journalists write in a very compact style, the result will almost always look decent. You just follow a few simple rules like:

  - Length of sentence without stopwords
  - Distribution of words across the article
  - Words shared with the title
  - Position in the article
  - Position in the paragraph
  - Does the sentence address the subject in third person? (avoid)
  - Does the sentence contain direct speech? (avoid)
Stuff like this. It's a bit of a Mechanical Turk, really.

There are no comments on this comment yet.