EDIT: More precise titles.

The way we sort comments on lemmy today is through giving everyone the ability to up and downvote any comment. You upvote comments you want to be more visible and downvote comments you want to have less visibility.

I think ranking comments are great if some opinions contribute more.

Imagine the value of a comment can be determined before voting. Perhaps with a panel of judges. With rankings -10 to 10 where -10 is dangerous misinformation, 0 has no meaningful contribution and 10 is perhaps mindblowingly enlightening or empowering.

Example of lemmy style sorting

Lets say a post is submitted. On the first day, it received the comments with a judges score (user, score):

(monkey, 8), (zebra, -2), (horse, 5), (panda, 3), (rhino, 3).

A post get the most visibility the first day. Therefore, those who post early can get a lot of votes.

However on the second day, one user submitted the comment:

(Flower, 9)

A this point in time the post doesnt have that many voters. And the top comments has increased visibility, so for every vote flower gains, the top comment might have received another. So we end up with the final ranking:

Monkey, horse, panda, flower, rhino, zebra.

So it ends up only ranking similar to panda and rhino even though the score was much higher. There was a mismatch between votes and score.

Solution, the topoligical sort

We should concider moving away from voting on comments individually to voting comparatively. Where you perhaps determine the most valuable comment out of a selection. Then a topological sort can put the best comment on top.

So even if flower is late to the party, their insight is still spread.

  • smallcirclesM
    link
    fedilink
    3
    edit-2
    2 years ago

    This is an interesting idea.

    Note btw that the default sorting for comments is ‘Hot’, which on first glance has some algorithm to it. In the screenshot below you see that the 2 days old comment with 3 upvotes is ranked higher than the 3 days old comment with 4 upvotes, and another 2 days old comment is below that with only 2 upvotes. The top comment here is non substantial (saying only “Thank you”) and the 2nd comment (also not very substantial) has replies to the discussion.

    Screenshot of some comments to a lemmy post.

    Hacker News

    Where I get most inspired where it comes to link aggregators…

    The Hacker News link aggregator has, I think, a very sophisticated set of algorithms to it that have been honed over many years. There’s more than meets the eye here in the deceptively simple app layout. Unfortunately they are not publicly known (only guessed).

    First of all a submission first exists in a separate site section for new submissions. It has to rise to the front-page, which requires a varying amount of upvotes which depends on amount of new links posted and the popularity (activity) of existing top-level entries. If activity is high enough, then it is entering the top-page immediately, or otherwise it might be added as new entry on the 2nd page where it has to ‘proof its value’ to enter top page.

    Once a submission is on the front-page it receives most attention (most members only ever check the 30 front-page entries). Now comments to a submission are not just top-ranked based on upvotes, but on multiple criteria. Like replies to a comment and the depth of the thread that is forming. And also on whether comments are substantive based on amount of text, and possibly also on keyword matches, URL’s posted, etc.

    When you submit a new top-level comment the place where it is injected also depends on multiple criteria. The activity on other subthreads and number of upvotes / downvotes in these trees. If your comment is substantive then it will stick on top of all other comments for a period of time, so it gets attention. The duration it sticks here depends on upvotes it receives. After that duration it starts being demoted in the list of top-level comments, but can rise again if there are popular replies.

    There are many more rules under the covers. For instance voting rings are detected, too many downvoted entries are ‘cooled down’, political heated discourse is downranked (is against the rules).

    I probably described some stuff inaccurately above. For more info see: List of Hacker News’s Undocumented Features and Behaviors.

    Your idea

    Where I think your idea brings difficulty is in the intuitiveness that is lost and friction added by doing something that’s more complicated than a simple up or downvote. You should not underestimate this. Who are the judges and how is their ranking determined? Note that the topological sort can also take place via weights that are calculated for an vote at a particular moment in time and depending on the entire state of the post.

    Lastly an interesting resource re:disinformation is the Bonfire Zappa project that Doug Belshaw is investigating from a user perspective.

    • Vegafjord eoOP
      link
      fedilink
      32 years ago

      The judges are not part of the alternate solution I suggested. They were only used for the sake of the example to quantify the value of the comments before anyone had voted. The judges ranking reflects the relationship between comments in terms of quality.

      The solution I suggested is one where voters compare comments and pick the best instead. Afterwards a topological sort is done to let the best float to the top. With enough comparisons, the ranking would be accurate.

      • smallcirclesM
        link
        fedilink
        12 years ago

        I see. I’d need to see how this would look like in the UX. For instance how can you compare comments if there are sub-trees of conversation attached to it?