Stories
Slash Boxes
Comments

SoylentNews is people

posted by on Friday August 25 2017, @11:26PM   Printer-friendly
from the we're-really-big-time-now dept.

Welcome, new trolls! We're pleased as punch to have you aboard, unfortunately as you may have noticed our moderators are unable to give you the moderations you've been working so hard for. Since we can't really do much about people not moderating more, we're going to be giving out more points so that the ones that do can give you the attention you so desperately crave.

Moderators: Starting a little after midnight UTC tonight, everyone will be getting ten points a day instead of five. The threshold for a mod-bomb, however, is going to remain at five. This change is not so you can pursue an agenda against registered users more effectively but so we can collectively handle the rather large uptick in anonymous trolling recently while still being able to have points remaining for upmodding quality comments. This is not an invitation to go wild downmodding; it's helping you to be able to stick to the "concentrate more on upmodding than downmodding" bit of the guidelines.

Also, this is not a heavily thought-out or permanent change. It is a quick, dirty adjustment that will be reviewed, tweaked, and likely changed before year's end. Questions? Comments?

 
This discussion has been archived. No new comments can be posted.
Display Options Threshold/Breakthrough Mark All as Read Mark All as Unread
The Fine Print: The following comments are owned by whoever posted them. We are not responsible for them in any way.
  • (Score: 0) by Anonymous Coward on Saturday August 26 2017, @04:01AM (10 children)

    by Anonymous Coward on Saturday August 26 2017, @04:01AM (#559275)

    Is anyone here thinking about setting up a reputation system? I've wondered for a while if that could improve internet discussions. Maybe based on hidden variable Bayesian networks, or some other modern technique to allow for different attitudes while still aiming for quality. Straight up/down voting seems so one dimensional.

  • (Score: 3, Insightful) by Azuma Hazuki on Saturday August 26 2017, @05:05AM (5 children)

    by Azuma Hazuki (5086) on Saturday August 26 2017, @05:05AM (#559300) Journal

    The problem with that is it's vulnerable to abuse in the long term. Registration is free and it's not hard to see someone with an axe to grind making a dozen burner accounts and Bayesian-poisoning a target to hell and gone.

    --
    I am "that girl" your mother warned you about...
    • (Score: 0) by Anonymous Coward on Saturday August 26 2017, @06:05AM (1 child)

      by Anonymous Coward on Saturday August 26 2017, @06:05AM (#559314)

      It would have to be resistant to that, so accounts with little history don't carry much weight.

      • (Score: 0) by Anonymous Coward on Saturday August 26 2017, @09:30AM

        by Anonymous Coward on Saturday August 26 2017, @09:30AM (#559360)

        Doesn't work - the attacker just makes noisier sock puppets, or lets them sleep longer before using.

    • (Score: 2) by NCommander on Saturday August 26 2017, @08:49PM (2 children)

      by NCommander (2) Subscriber Badge <michael@casadevall.pro> on Saturday August 26 2017, @08:49PM (#559581) Homepage Journal

      This is primarily the reason metamoderation was never deployed to production. M2 (as Slashdot implemented it) was fixed almost two years ago, but I thought the concept was fundamentally flawed based on our experience with how people moderate on SN.

      --
      Still always moving
      • (Score: 2) by Reziac on Sunday August 27 2017, @12:56AM

        by Reziac (2489) on Sunday August 27 2017, @12:56AM (#559660) Homepage

        I meta-modded a few times back when it was new, then decreed it a waste of time and never did it again.

        --
        And there is no Alkibiades to come back and save us from ourselves.
      • (Score: 0) by Anonymous Coward on Sunday August 27 2017, @02:37PM

        by Anonymous Coward on Sunday August 27 2017, @02:37PM (#559853)

        The /. version had issues. I think adding an incorrect good or incorrect bad mod option that has no points associated with it could possibly be used to help evaluate whether people are too often modding things wrongly.

        But, that would be a tremendous pain to get working right and shy of having people actually looking at the moderation that's being done, it's probably always going to be a problem.

  • (Score: 4, Insightful) by tibman on Saturday August 26 2017, @05:05AM

    by tibman (134) Subscriber Badge on Saturday August 26 2017, @05:05AM (#559301)

    Could we call it something cool like.. Karma?

    --
    SN won't survive on lurkers alone. Write comments.
  • (Score: 2, Interesting) by khallow on Saturday August 26 2017, @07:22AM (2 children)

    by khallow (3766) Subscriber Badge on Saturday August 26 2017, @07:22AM (#559331) Journal

    Maybe based on hidden variable Bayesian networks, or some other modern technique to allow for different attitudes while still aiming for quality.

    I would suggest singular value decomposition. (Sorry, but I'm using a bunch of linear algebra lingo in this post.)

    Consider a model of users' opinions of post quality versus the posts. If somehow each user were to completely record mods for every post ever, it'd be a rectangular grid with a lot of numbers (rows being users, let's say, columns being posts, and individual cells being the opinion value for a particular user and post). One can approximate that grid as a numerical matrix of the form U*D*V, where U and V are orthogonal matrices with respect to the far left for U and far right for V. D is a diagonal matrix of weights or "singular values". When D has the same rank as your original matrix, then the approximation is perfect, up to round off error.

    But since you don't want the full information, you can throw away all but a fixed number of the highest weight singular values and keep only the parts of U and V that multiply by those weights.

    Basically, it's very close to an eigenvalue/eigenvector approach which it reduces to, when the matrix is square (V would be matrix of eigenvectors, D matrix of eigenvalues, and U the adjoint of V).

    Anyway, this method also lends itself to filling in huge gaps since in the real world, we're not going to mod every post possible. But one can calculate U,D,V triples that approximate moderately well the mods that have been made.

    Pros:

    1) allows for several independent modding dimensions rather than the one of the current system.
    2) can impute a mod for any user or post with sufficient mods (say 5-10 mods made by the user or put on a post).
    3) Given a user, can estimate a mod for any post with one or more mods.
    4) Insensitive to usual mod bombing and such. Even a large number of users with the same opinion would just look like a single dimension to the algorithm. Mod poisoning is harder.
    5) iterative method so it can be updated with more recent changes in opinion.

    Cons:

    1) Complicated algorithm with significant overhead and some unknown interactions.
    2) I can't figure out a way to do imputation in real time. I'm thinking the U, D, V matrices get computed every fixed unit of time (say once a day, for example, with the previous day's U, D, V matrices used as initial guesses). You'd also want the number of mods to be limited in some way to reduce computational effort.
    3) Dimensions don't have specific meaning (they're just dimensions along which the mod matrix would be best approximated with a fixed low rank matrix) while the dimension of the current mod system does.
    4) For stuff, you haven't explicitly modded, you would see drift in the mod estimates with each update.

    5) Each page load would including mod calculations on each post specific to the user making the page request. Could be a lot of overhead to page generation and loading.