That real question is, what problem are we trying to solve? Then we can go from there.

What we have right now in Lemmy strikes the current balance IMO. Individual comments are upvoted/downvoted. But no cumulative score.

Here’s a crazy idea. What if down voting a comment/post resulted in a weighted random float between 0-1 while upvoting resulted in a weighted random float between 1-2? If you virulently hate a comment or post, ignoring it is the surest way to bury in completely. Posts and comments that Garner attention become the most visible, but gaming the system for visibility could become difficult if the weighting algorithm was tuned appropriately.

Score the posts, not the individuals. Attaching imaginary points to any kind of activity instantly turns it into a competition.

Instead, any scoring should focus on actual content, which is basically what the up/down vote is.

We should keep it as is. Having an account score just amplifies a big issue with sm. The content should be in focus, not the people posting. A relevant comment should be hightened because it itself is good. In the same way we shouldn’t judge something because the user has a low karma, but because the content is bad.

The idea behind something keeping a score on a profile is good, but it doesn’t work as intended in practice. People will farm in whatever way they need to get a moral highground. Not having such a scoring system will be a good way to reduce the incentive to copy/paste content from others.

Web of trust. The biggest thing missing from most attempts to build social networks so far. A few sites did very weak versions, like Slashdot/s friend/foe/fan/freak rating system.

Let me subscribe, upvote, downvote, filter, etc specific content. Let me trust (or negative-trust) other users (think of it like “friend” or “block”, in simple terms)

Then, and this is the key… let me apply filters based on the sub/up/down/filter/etc actions of the people I trust, and the people they trust, etc, with diminishing returns as it gets farther away and based on how much people trust each other.

Finally, when I see problematic content, let me see the chain of trust that exposed me to it. If I trust you and you trust a Nazi, I may or may not spend time trying to convince you to un-trust that person, but if you fail or refuse then I can un-trust you to get Nazi(s) out of my feed.

I asked chatgpt, just to see what would filter out…

I would like some advice on designing a content-sorting and content-filtering mechanisms for discussion forums, in order to avoid or mitigates some of the problems that current systems are prone to.

One of the social problems with discussion forums results from the concept of upvoting or downvoting content.

It can be argued that voting content up or down serves to sort content by quality, allowing high-quality content to be seen and engaged with, while discouraging low-quality submissions.

However, in practice, Goodhart’s law ends up applying: when a measure becomes a goal, it stops being a useful measure.

By using popularity as a proxy metric to determine quality, and by rewarding popularity with positive attention, this ends up selecting for content with superficial emotional appeal - ragebait, memes, facile/obvious comments pandering to common sentiment, puns, etc. - and not ‘useful’ content that is thoughtful, incisive, analytical, or important-yet-unpalatable. Ironically, content of this nature is also low-quality in a different way.

Worse, this ends up training users to produce content of this nature at the expense of thoughtful, interesting, incisive discussion, reducing both the production of quality content, and the quality of the collaborative sorting that users perfom by way of voting.

A cumulative ‘karma’ score for users, being a sum total of their upvotes and downvotes across all of their submissions, while ostensibly encouraging engagement and admirable behaviour… can again end up being gamified, and lead to users making large numbers of popular but low-effort submissions in order to maximise their score.

This can also produce ‘filter bubbles’ or ‘hive minds’: subcultures or communities that reject and discourage disagreement or criticism, and end up with a positive feedback loop leading to increasing orthodoxy with little grounding in reality. In some cases, the communities involved can become radicalised or toxic as a result, as more-moderate voices are suppressed from the discussion, and increasingly-extreme views become required in order to gain attention. Of course, it’s reasonable and useful to let groups based around common ideals preserve an individual identity separate from the mainstream of discussion, but perhaps a way to regulate excessive self-selection would be useful.

Doing away with any kind of quality-sorting mechanic altogether is not a very useful alternative; a forum where spam, antisocial, or irrelevant submissions are given equal weight and visibility as interesting or thought-provoking posts tends to discourage users from participating or even reading.

As a further complication, users do not generally follow instructions or policy set out for voting on posts, and will tend to use them to indicate approval/agreement regardless.

Human moderation is an important part of any discussion forum and essential to the guidance ans shaping of any community, but the workload does not scale well; it should be considered a limited and intensive resource.

It would seem there are a few challenges to overcome:

1: Better-define the concept of ‘quality’ for discussion-forum posts beyond immediate, superficial emotional appeal - what is it that makes discussions feel satisfying, and drives deeper engagement? 2: Find a way to programatically ascribe a metric based on this concept, via the engagement of other users, in any form including upvotes, downvotes, tagging, commenting - and statistical or derived data such as the number, depth and overall score of replies to the post, and perhaps even analysis of the text of replies. 3: Ensure that any such system is difficult for individual users to exploit or game; for instance a naive reply-count metric could be trivially exploited by a user posting a large number of replies to a single comment.
4: Limit the effect of self-selection feedback loops, to prevent radicalisation while still allowing common ideals and identity to flourish in their own spaces.

One idea I’ve had would be to sort all content by ‘controversialness’ - that is to say, posts with a high number of total votes, but a low-magnitude score once downvotes are subtracted from upvotes. I believe this would tend to deprioritize low-effort posts such as ragebait or pandering to common sentiment, along with junk, spam and boring or irrelevant content, and thus bring ‘interesting’ content to the top. I believe it would be difficult to game for attention, or for other users to have undue effect on its visibility. However, I’m not sure whether controversialness would be a good metric to drive engagement, or what large-scale effects sorting on this metric might have on a community.

Can you offer any advice or commentary on the situation, or identify any important flaws in my analysis? Can you suggest any alternatives that might work better?

===================================================================================== ChatGPT You’ve outlined several important challenges and considerations when it comes to designing content-sorting and filtering mechanisms for discussion forums. The issues you’ve raised are indeed prevalent in many online communities, and finding effective solutions can be complex. I’ll provide some advice and commentary on your analysis, as well as suggest alternative approaches that might address the problems you’ve highlighted.

Better defining quality: You’re right in emphasizing the need to define quality beyond superficial emotional appeal. One way to approach this is by incorporating multiple dimensions of quality, such as insightful analysis, thought-provoking content, well-reasoned arguments, and valuable information. Encouraging constructive and meaningful discussions should be a primary goal.

Ascribing metrics based on user engagement: Rather than relying solely on upvotes and downvotes, consider implementing a more comprehensive metric that takes into account various aspects of user engagement. This could include factors such as the number of replies, the depth of discussion, the length of time spent reading the post, and even sentiment analysis of the comments. By incorporating a broader set of metrics, you can obtain a more nuanced understanding of user engagement and use it to evaluate content quality.

Avoiding gaming and exploitation: To mitigate the risk of individuals exploiting the system, it’s crucial to design mechanisms that discourage or penalize excessive or low-quality contributions. For example, you could implement rate-limiting measures that prevent users from making an excessive number of submissions within a certain time frame. Additionally, considering the quality of the engagement (e.g., the relevance, thoughtfulness, and contribution to the discussion) rather than just the quantity (e.g., sheer number of replies) can help minimize gaming attempts.

Addressing self-selection feedback loops: To prevent echo chambers and radicalization, it’s important to strike a balance between allowing communities with common ideals to flourish and promoting diverse perspectives. You could introduce mechanisms that encourage exposure to alternative viewpoints, such as algorithms that prioritize showing posts to users with different preferences or even implementing mandatory or randomized exposure to different topics or communities. Additionally, fostering a culture of respectful and open-minded discussion can help create an environment where diverse opinions are valued.

Regarding your idea of sorting content by controversialness, it could be an interesting approach to promoting engaging discussions. However, it’s important to carefully consider the potential effects of such sorting on the community. Controversial topics and discussions can be divisive, and if not managed well, they may lead to increased conflict and hostility. Implementing safeguards, such as active moderation and community guidelines, would be essential in maintaining a healthy and constructive environment.

In conclusion, tackling the challenges of content sorting and filtering in discussion forums requires a multifaceted approach. By defining quality beyond superficial appeal, incorporating comprehensive engagement metrics, preventing gaming and exploitation, and addressing self-selection feedback loops, you can design a system that promotes meaningful discussions while minimizing the negative effects associated with current mechanisms. It’s crucial to iterate and refine these systems based on user feedback and community dynamics to ensure the best possible experience for forum participants.

I like the system as it is here at the moment. Up-/Downvotes per Post/Comment to show the popularity (and express (dis-)approval). But nothing to collect per account, so noone gets encouraged to post just for the karma.

Posts should just be upvoted and downvoted with no credit given to the person who posted. Same goes for comments. In my opinion, upvoting and downvoting should just help the user find the most relevant information. Content that people upvote is the most seen. Content that people downvote is the least seen. Posters and commenters stay on an equal footing with no points system.

I’m against any kind of global user ranking.

It makes sense to rank content, but ranking users just begs abuse of the system. There’s always those that will try to farm the system resulting in lower quality content. It’s also an attack vector for bots.

I don’t miss the “karma” aspect one bit here. Rate my post quality, not me. On the other hand, tools for ranking users privately could be helpful. In other words a personal ranking for your eyes only would be fine.

Advogato reputations: https://en.wikipedia.org/wiki/Advogato#Trust_metric

I like it as it is to be honest.

Karma and votes should stay but be hidden to other users. Karma is a good way to detect bots and trolls, but just admins and moderators should see it to act on them if needed. And up/down votes should be hidden too because of the hive mind phenomenon that it produces (Experienced on Reddit): often, the funny or sassy or apparently clever comment gets upvoted and sometimes, the comment with knowledge about the post gets downvoted because the first joke was funny. Many people may not have an opinion about the issue but upvote the funny guy and downvotes the real answer just following the hive. Hiding it, each person reading must decide by themselves if they upvote or downvote a comment.

Prizes and awards could maybe stay, not sure

What about hidden karma?
Like there is still karma used internally to decide what posts to promote and how to weight votes, but the numbers are kept only internally so people don’t get obsessed with that number next to their (and others’) profile?

I still firmly believe one of the worst things to happen to the internet, besides pop-up ads, is up and down votes. Nothing exposes a misanthrope quicker than forcing them to comment instead of passively downvoting everything they see. Which makes it easier to remove them from the party.