Twitter  This Week in Review: Facebook thinks they have the formular for cracking down on clickbait nie.mn/1thp0Ds  
Nieman Journalism Lab
Pushing to the future of journalism — A project of the Nieman Foundation at Harvard
Horse with jaunty gallop

How a New York Times developer reverse engineered @Horse_ebooks –An Interesting

Journalists should always be hacking, trying to tell stories in surprising new ways.

Horse with jaunty gallop

Jacob Harris, a senior software architect at The New York Times, shares my obsession with @Horse_ebooks, the wise and mysterious Twitter spambot. @Horse_ebooks tweets nonsensical phrases, apparently scraped at random from the web, and sometimes includes links to spam sites. The account has become a huge hit, with 74,000 followers.

The horse is often imitated but never duplicated, powered by the manual labor of human satirists. Like a good hacker, Harris took his obsession to the next level: He reverse-engineered the horse’s algorithm and created one for the New York Times. Behold, @nytimes_ebooks:

Like its precursor, tweets from @nytimes_ebooks are surprisingly compelling and accidentally hilarious. Harris describes in a blog post how he did it: A script crawls the New York Times RSS feed for recent stories, extracts quotes from the text (“better for ebookification,” he writes), and converts the text into a Markov chain.

Harris has no control over the text produced by his bot, which he finds “both comforting and alarming.” The source material includes the darkest moments of the human experience. He said the project is not unlike the artwork in the Times’ 8th Avenue building, a series of mounted screens that pluck phrases from stories and flash them without context.

Unlike its precursor, every @nytimes_ebooks tweet includes a short link back to the story. And it’s nearly impossible to resist clicking to find out what inspired the nonsense.

Bravo, Jacob Harris. You probably generated enough clicks per reader to hit the NYT paywall many times over.

“There is a mystery in the Markov model of how it writes its text,” Harris told me in an email. “Like Eliza or other textual experiments, there is this ambiguity where the machine sometimes writes something poetic and new and sometimes line noise. If this were a 3-D drawing of a person, we’d be staring right at the horror of the uncanny valley, but here it’s really compelling. Why?”

A father of two young children, these are the kinds of thoughts he discovered in the “loopy predawn hours.”

“Sometimes the text reminds me of a toddler learning to talk. We like to watch it because sometimes bots say the darndest things! And sometimes because it feels like we’re watching something being born.”

Harris laid out an example:

I’m not sure if @horse_ebooks uses this attention to get clicks. I’ve never clicked on a link in its feed when they appear. But I could see how you might want to just to see where the tweet came from. For instance, here are two tweets of the same story.

Which would you click? Of course, I did this for the lulz, not the clicks, but I’d be interested to see if it has a positive effect there, given that it’s not user-friendly at all! To give you some background, The New York Times sometimes creates two headlines for an article: a print version which can be opaque and artful and a more straightforward version of the headline for mobile readers and twitter. This makes sense, because print readers can see what the article is about from its context and layout on the page, but a headline like “A Very Fine Line” would be opaque and annoying on Twitter (where it ran as “A Brooklyn Artist Free-Associates on Her Walls”).

Harris stresses this is nowhere near an official project of the Times. But this being the Nieman Lab, we try to take away lessons for the news business. @nytimes_ebooks demonstrates the joy of finding content in unexpected places, places that previously appeared to have none. Who would have thought a robot that slipped through Twitter’s spam filters would have inspired so much creativity? Content with limited value in one context can have real value in another.

It’s a great example of the hacker mindset that journalists can embrace: What is a truly new and surprising way to tell stories? Experiment often, fail fast. Harris told me he spent a few days tinkering with Markov chains and two evenings coding it, but that’s it.

Are you confident that,

                                   
What to read next
Quartz_homepage
Joseph Lichterman    Aug. 26, 2014
Previously proudly without a homepage, the business site is trying to shift its email success to the web to build loyalty.
  • http://chesmart.in ches

    Cool story, but — forgive me being pedantic — this not reverse engineering. This is simply creating a new bot using Markov chains, a well-known process for generating text like this.

  • http://andrewphelps.com Andrew Phelps

    You are correct; Harris did not reverse engineer the algorithm per sé. He just figured out a way to write a program that automatically generates tweets that look a lot like the Horse’s, whereas I think a lot of imitation accounts are done by hand.

    I, for one, had never heard of a Markov chain.

  • http://twitter.com/harrisj Jacob Harris

    But now you can’t stop thinking about them, can you?

  • http://andrewphelps.com Andrew Phelps

    CAN T STOP THINKING ABOUT