back to article Yahoo! Japan! offers! free! comment!-moderation!-as!-a!-service! API!

Yahoo! Japan has flicked the switch on an AI-driven comment-moderation-as-a-service API. The independently owned outpost of the famed web portal has used a homegrown natural language processing model to keep the comments section of its own News service clean since 2007, and claims the AI deletes 20,000 posts per day. The tech …

  1. Chris G

    By any other name

    This is a system of censorship that seems to be used to promote a narrative or theme.

    While the need to vet and moderate comments is both useful and necessary, the article's description of this moderation Algorithm gives me the impression that it is being used to direct conversation rather than encourage open discussion.

    1. Pascal Monett Silver badge

      Re: By any other name

      I'm sorry but I do not come to that conclusion after reading this article.

      Especially since this paragraph is in rather direct contradiction with your words :

      The tech, styled as a "constructive comment ranking model", prioritises posts that provoke positive discussions, especially when they suggest new ideas or are insightful. Comments that mention experiences related to articles also do well.

      What I do think is that, even if this API works, it will only work in Japanese, and I tend to not comment in that language, for some reason.

      I'd like to see someone port that to English and implement it on Twitter.

      The silence will be deafening.

      1. Chris G

        Re: By any other name

        " prioritises posts that provoke positive discussions"

        It all depends on who decides what the priorities are and what constitutes 'positive'.

        There are many publications that use the 'Does not conform to community guidelines' as a takedown mechanism for anything they dislike or that conflicts with their particular stance.

        I presume there must be a way to set parameters, otherwise how can it work for each and every use?

        The three different organisations mentioned in the article can hardly be using exactly the same parameters and Yahoo Japan is zapping 20,000 posts a day which seems a lot to me.

        1. Jimmy2Cows Silver badge

          Re: depends on who decides what the priorities are and what constitutes 'positive'

          Precisely this ^^^

          The differences between a controversial viewpoint promoting healthy discussion of a topic, and being outright trolling, may be subtle. People can struggle to spot that at the best of times, never mind if handed off to an AI which may or may not have been trained to meet someone's specific agendas and personal / corporate takes on morality, ethics, offensiveness, triggers and positivity.

        2. Dan 55 Silver badge

          Re: By any other name

          Yahoo Japan is zapping 20,000 posts a day which seems a lot to me.

          I'm surprised it's only 20,000 posts a day.

      2. bombastic bob Silver badge
        Big Brother

        Re: By any other name

        The silence will be deafening.

        THAT would be the ONLY positive outcome possible. Yet HACKERS will always be able to cheat any algorithm. I suppose a proper set of snarky adjectives and twisted metaphors would be a good start...

  2. Francis Boyle Silver badge

    constructive comment ranking model vs amanfromMars

    Showdown of the century.

    1. ConsumedByFire

      Re: constructive comment ranking model vs amanfromMars

      Hilarious. I used to spend time trying to elicit meaning from some of the more surreal posts but theres only one life.

      I would be interested to see what the algorithm made of Bombastic Bob (who I sometimes agree with sometimes don't but always appreciate his forthright approach to commenting).

  3. 2Blockchainz

    A natural experiment

    Yahoo! Japan's continued existence and profitability demonstrate serve as a good comparison to their US brethren managerial incompetence through the decades:

    Verizon, Marissa, Bratz, Yang, (particularly) Terry Semel

  4. Anonymous Coward
    Anonymous Coward

    Fox in the hen house

    Letting Yahoo moderate user comments?

    Have you ever seen the user comments of Yahoo news?

    It would make users of the Parlor app blush.

    Some of the most racist, hate-filled crap I've ever seen.

  5. NicX

    Neat in theory, not so much in practice

    This isn't going to remove a lot of things it should, and will promote a lot of things it shouldn't. Just the nature of the beast.

    The only use case I can think of for this is racist or overtly hateful comments. Other than that, let adults be adults and manage their sensitivities. If reading a...foul...internet comment rattles you that much, you may have some other issues you need to work out.

    1. bombastic bob Silver badge

      Re: Neat in theory, not so much in practice

      A carefully worded comment might illustrate how you might easily get past an AI algorithm.

      (but I'll leave that one as an exercise)

      This also reminds me of that old phrase, "the tail wagging the dog" when "the few" (easily triggered) must control "the many" (who lose freedom).

      So humans with soft-touch moderation are needed to fight off the trolls and bots. However, an AI is more likely to behave like an aggressive spam filter, where e-mail from your mom is marked 'spam', but e-mail from scammers and 'male enhancement' vendors get through. EVERY! SINGLE! TIME!!

      The internet does not HAVE to be a sewer. But it is. Maybe a click-through disclaimer is needed?

  6. Anonymous Coward


    Moderation algorithms, by their very nature, are meant to be biased. What they are biased against is the question most people debate. I suspect that discussing the royals in Japan is an algorithmic minefield.

    A better idea would be to restrict comments on generic news sites. Does anyone need to see my views on the vaccination passport that is Yahoo Japan's top story at the moment? Does anyone care? Ditto the Yahoo US story on the Arizona recount or the UK story on surge testing?

    Of course ElReg is anything but generic so we should be fine. ;-)

POST COMMENT House rules

Not a member of The Register? Create a new account here.

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon

Other stories you might like