Slashdot is powered by your submissions, so send in your scoop

 



Forgot your password?
typodupeerror
Get HideMyAss! VPN, PC Mag's Top 10 VPNs of 2016 for 55% off for a Limited Time ×
AI Technology

Machine Learning Generates Clickbait Headlines That Will Shock You! (thestack.com) 100

An anonymous reader writes: Norwegian developer and blogger Lars Eidnes has designed a clickbait generator using a neural network, which is able to create sensationalist headlines that play on human readers' curiosity. Eidnes trained his neural network by scanning around two million clickbait titles from online media sites such as Buzzfeed, Jezebel and Upworthy. When asked to form a sentence, the system can now output a single word and continues the prediction process to find related words, in a pattern known as Recurrent Neural Networks (RNNs).
This discussion has been archived. No new comments can be posted.

Machine Learning Generates Clickbait Headlines That Will Shock You!

Comments Filter:
  • by sims 2 ( 994794 ) on Friday October 16, 2015 @12:31AM (#50741269)

    Annoy is more likely.

    • The text makes about as much sense as the typical over dramatic Facebook post everyone sees constantly. It would be funny to train the NN with posts of that sort, and generate entire Facebook "personalities" that blend in with any other angst-ridden grammatically-challenged Facebook poster. Maybe it would even confuse the data scrapers and spies out there, who knows?

      It ought to be funny in any case.

      • by Gr8Apes ( 679165 )
        I think they've already done this. On a side note, all those headlines do is train you to ignore them, which in itself could be somewhat dangerous. Better to ignore the sources generating them instead.
    • What I want is a WWN clickbait generator: Woman gives birth to two-headed fish, WWII bomber found on moon, Cat owns 23 old ladies, Reporter eaten alive by rabid hamster, Bat child found in cave, yeah, now that's clickbait.
  • by fragMasterFlash ( 989911 ) on Friday October 16, 2015 @12:31AM (#50741271)
    How do I downvote this headline?
  • Took the bait. (Score:5, Insightful)

    by AndyKron ( 937105 ) on Friday October 16, 2015 @12:32AM (#50741275)
    Ya, I clicked on it.
  • by Anonymous Coward

    Machine learning programmers everywhere hate it!

  • on /. about sports and financial writers being replaced by computer programs? Seems anything with a lot of statistics was ripe for this since sports/finance writers were really just putting fluff around numbers. Good thing automation never [cnn.com] costs jobs, right?
    • by Anonymous Coward

      If you want an example check out https://www.google.ca/finance?q=sial
      Sigma-Aldrich was bought out by Merck over a year ago for $140/share, but they had to wait for approval from various countries, "with the approval of the European Commission (EC) being conditional upon the sale of parts of Sigma-Aldrich's solvents and inorganics business in Europe" which they're now waiting to close. Until that happens there's very little interesting financial news about the company, and the share price has been steady a l

  • It's still not better than man-made Engrish.

  • by Greyfox ( 87712 ) on Friday October 16, 2015 @12:40AM (#50741303) Homepage Journal
    Come with me and you'll be
    In a world of headline automation!
    Take a look and you'll see
    headline automation!
  • From yesterday's Slashdot entries:

    FBI and Join UK Against Forces Against Spread of Dridex Banking Malware [slashdot.org]

    Sounds like machine-generated gibberish to me.

  • the URL's next to the article posting are!

  • Obligatory (Score:3, Interesting)

    by pushing-robot ( 1037830 ) on Friday October 16, 2015 @12:59AM (#50741345)

    When asked to form a sentence, the system can now output a single word and continues the prediction process to find related words

    https://xkcd.com/1427/ [xkcd.com]

    Come to think of it, I need a browser plugin that randomly cuts off the end of sentences and replaces them with autocompleted words.
    Extra credit if the algorithm was trained on religious texts, trashy romance novels and operettas.

  • AI My Ass (Score:5, Informative)

    by Jane Q. Public ( 1010737 ) on Friday October 16, 2015 @01:01AM (#50741357)
    It may have been generated via "AI" techniques, but the results are little if any better than a simple Markov-Chain text generator, of the kind I built back in the 80s with my programmable calculator... and used today to generate spam emails.

    The only difference I see is that there are pictures accompanying the "articles". But I don't see anything about an algorithm for that, so I think I am safe to presume that the text is generated by the so-called "AI", but composed by an actual human.

    But maybe not... the pictures seem pretty random and unrelated to the texts. So it could be done algorithmically... but then it's still just as UNimpressive.
    • Before anybody misunderstands... I am not using them to generate spam emails. But you've most likely seen some.
    • ... And here is some output from a simple (naive) Markov-Chain generator fed Alice in Wonderland:

      It unfolded that it and then all the water and see whether it's marked 'poison' or not," for she had nibbled some more tea," the Cat said, without opening itself up on tiptoe and peeped over them, all ornamented with her head through that day.

      "You ought to her lips, saying tone; "don't be nervous, or I'll have finished," said the King, "call the party I ever was a little feeble, squeaking voice.

      "Then you should say what was coming to, but it was all ridges and found that it led into a tidy little cart-wheels and that's a fact."

      Just then Alice (she had kept a piece out of sight.

      Alice was a paper label, with the wood. "It's a Cheshire-Cat," said the Cat in a languid, sleepy voice.

      "What do you know about the same solemn tone, "For the fire, stirring a long hookah and take it and there. There was no "One, two, the Caterpillar took the house before seen a rabbit with the Queen, stamping about something; Alice heard this, she found that sort.

      Next came the royal children; there were trying every now and then hurried out, "The race was over at the Mouse only shook its head impatiently and was just begin?"

      This question is 'What?'"

      Alice was just going to get very tired of swimming away.

      "You're looking about for serpents! There's no name signed at the same thing, you know about two feet high and was "Why is a raven like a writing-desk?"

      "I'm glad they'll do next!" thought, "it's sure to do with you. Mind now!"

      "He took me for her neck from being broke off and Alice heard it muttering to eat or drink under the immediate adoption of more energetic remediesâ""

      "Speak English!" said the Pigeon, raising its eyes, "Of course," said the King and very gravely, "I think of anger, and tried to her feet, for it flashed across her mind that she did not get hold of any good reason to be afraid of it.

      Presentâ"at least notice this morning I've nothing to do.' Said the Knave of Hearts.

      The King looked up and bread and birds with draggled feathers, the animals with the words "DRINK ME" beautifully printed on going to turn into a cucumber-frame or some way out of prisoner toâ"to somebody to talk to."

      "How are you grow shorter, until there was hardly knowâ"No more, thank ye. I'm better now," Alice thought the flamingo and tucked it and people up like a telescopes. There was a general chorus, "Yes, please do!" But when they had been. world of trees, and I've tried her way through was more hopeless than ever.

      Parameters are adjustable. It can be made to make more sense than this, and there are some refinements to the general technique that aren't present here, particularly in the way of punctuation. I just generated a quick-and-dirty on the fly.

    • It may have been generated via "AI" techniques, but the results are little if any better than a simple Markov-Chain text generator, of the kind I built back in the 80s with my programmable calculator... and used today to generate spam emails.

      You sent spam emails today? Dickhead.

      (For the stupid: I know what he meant.)

    • "New Rule: Lawrence V. Williams, Turkey Or Child -- A Good Guide"

      If loving this is wrong, I don't want to be right :)

      • "New Rule: Lawrence V. Williams, Turkey Or Child -- A Good Guide"

        If loving this is wrong, I don't want to be right :)

        "Taylor Swift: The New Face of Victim of Peace Talks"

    • It may have been generated via "AI" techniques, but the results are little if any better than a simple Markov-Chain text generator, of the kind I built back in the 80s with my programmable calculator... and used today to generate spam emails.

      The only difference I see is that there are pictures accompanying the "articles". But I don't see anything about an algorithm for that, so I think I am safe to presume that the text is generated by the so-called "AI", but composed by an actual human.

      Try reading the article. The pictures are selecting using a standard image search with the generated headline as topic.

    • by martas ( 1439879 )
      Neural network marketing is surprisingly good. After the initial wave a few decades ago of computer scientists promising to solve all the world's problems by reinventing basic statistical methods under sexy names, I thought the hype had mostly died down by the mid 2000's. But there's been a hell of a comeback, I think in large part because of very successful marketing by the likes of Andrew Ng and whoever is behind the deep learning stuff at Google.

      And then there's the recent wave of high profile doomsayi
    • Unless they also built an effective parser and include lexical information, this is probably just is an n-gram generator. It doesn't matter how 'deep learning' the neural network, maximum entropy, or naive Bayesian is; you get out what you put in the model.
  • The method Eidnes used is hardly anything new. Frankly, I'd be surprised if many existing clickbait headlines weren't generated in this way to begin with. After all, it'd be, I think, easier to run some statistical analysis on the headlines that get the most traffic, training the generator accordingly, than to try to conceive of catchy-sounding headlines for every piece.

    Still, it's obvious he put some effort into this, and it's somewhat interesting to read about, so at least there's that.

    • You (probably) joke, but the unsettling thing is that you're right. I think this has things to teach us about 'popularity' and 'survival of the fittest'.

      Perhaps I'm just crazy, but it seems to me that if this works in practice, it disproves 'market' theory. In other words, if you reduce all friction and gatekeeping, everything becomes shit :)

      This will come as a shock to many 'disruptive' startups believing they're making a better world aaah, who am I kidding? They're FINE with that as long as it makes them

      • Not everything becomes shit. Since these tricks don't work on you, you would be an untapped market using just them. All this proves is that you can sell shit to undiscerning people, which I think we already knew.
        • Are you kidding? I have a horrible fascination with US 'Kitchen Nightmares'. These tricks totally work on me :D

          There is NO market that isn't completely destroyed by the removal of friction and gatekeepers. It's shit all the way down!

  • Snowden Tweets Hillary in Linux: NodeJS Cows found on Pluto Goatse Beowulf Cluster of Hot H1B Grits!

    • You missed a few clickbait terms: Windows 10, Systemd, Women in Tech, SJW, Gamergate, 3D Printing, NSA, Network Neutrality, Comcast, Drones...

      I should probably be modded down -1 Flamebait just for the sheer volatility of combining all these terms into one sentence.

    • by msimm ( 580077 ) on Friday October 16, 2015 @03:25AM (#50741685) Homepage
      Too old school for most of the remaining readers. Slashdot, a site that initially attracted a high number of reasonably highly technically skilled readers, remains at the brink of irrelevance. That's a headline. Pandering to the lowest common denominator was cute back in the day when we all knew it was an inside joke. The typos and overt clickbate along with the inside jokes were fun. But after a while when a friend doesn't grow with the rest of his peers...eventually the ole hijinx get old. Nobody should want to be a 40 year old stoner journalist failure or to emulate one.
  • by phayes ( 202222 ) on Friday October 16, 2015 @02:07AM (#50741521) Homepage

    The name they gave to their invention: Timothy

    What a coincidence, we have our very own clickbait generating Timothy here on /. too!

  • And I think I will stop opening slashdot too...
  • Greenpeace feminists conspire with HB1 Visa holders to take guns away from experienced middle aged coders!
  • Stanislaw Lem's 1976 short story "One hundred and thirty seven seconds" ("Sto trzydzieci siedem sekund") is strangely similar to this (except for the fact that they eventually conclude the article-writing neural network can see 137 seconds into the future).

  • If the computer can generate headlines making you think of Betteridge's law of headlines, then journalism is really dead, that was the last bastion of the living, breathing journalist.

  • I am confused, What is supposed to shock me more? is it that this actually made it here as a story or how incredibly bad the AI works? I could write a random word generator with a few simple English language rules that would come up with more shocking headlines. The samples would not even be close to being clickbait on any site they are pathetic, if anything the headlines PRIOR to it learning were actually more likely to generate clicks as at least they have a "what the fuck does that mean" clickbait attach
  • How often can you see a headline that says "you'll be shocked" or "you won't believe" before semantic satiation sets in and it just becomes meaningless text that you scroll past? I don't think those phrases have raised any interested from me since sometime in the 90s.

  • Now they can let go their editorial staff and generate their site entirely by scripting
  • by Tom ( 822 )

    That is one step closers to automatically detecting this clickbait bullshit, so my ad blocker can filter it out...

    That's what they intend to do, yes? Right?

  • Because my cellphone has been doing this for years. All I have to do is bash the space key and I get tonnes of random poems and headlines generated on the fly.
  • Come on guys, I need an extension to Chrome and Firefox. It should open a pseudo tab, that is actually never rendered on screen. It should click on all the click bait links. Should have some AI to recognize clickbait headlines. The neural network trying to learn my click bait tastes should go completely haywire.

    Wish I have the energy to write one. But after hacking all day at work, I dont have the stamina to embark on more coding in my leisure. It should be childs play for those who hack browser extension

    • What I want is a simple "Click-bait" button on my browser and the Android Facebook app. When I click it, the site that had the "click-bait" headline is added to my personal [i]index expurgatorius[/i], and as far as my online experience is concerned, it utterly ceases to exist.
  • This will be a perfect testbed to train my clickbaitblock plugin to go along with my adblock plugin.

  • "If I Were People Inside The Way My Kid Was In Syria, I Would Find Next Child If. And I Will It Call Them 'Save Me'", "Lawsuit Claims United Building's North American Movies Are Over", "50 Shades Of Baby? President Bill Clinton Admits She's Ready To Want You Back"... This story will be more interesting when the software can filter out nonsense headlines or correlate the components of the headlines better than it is now. Good base idea for a project, though.

"It's like deja vu all over again." -- Yogi Berra

Working...