Updates to "Activity" sorting method (the site's default)
Since it launched, Tildes has always been using "Activity" as the site's default sorting method, which behaves like a classic forum—any new comment in any topic causes it to "bump" back to the top of the list. This has generally worked well overall, and has been a good way to keep threads visible and active over longer periods.
However, there have been a few issues related to it, such as controversial threads staying at the top of the site for long periods of time, and bickering back and forth between two users causing threads to constantly bump back up to the top even if nobody else is interacting with the topic at all. We haven't had great ways to deal with this so far, and have mostly had to work around it by setting the default time period to "last 3 days" so that threads can't dominate the site indefinitely, or even locking threads to force them to drop off.
As an attempt at a better solution, "Activity" has now had its behavior changed so that topics will only bump to the top when something "interesting" happens, instead of for every single comment. The exact methods we're using to determine "interesting" will need experimentation and to be adjusted as we see how they work, but initially it's entirely based on comment labels:
If a comment or any of its parent comments has an active Noise, Offtopic, or Malice label (note: it generally takes at least two users applying the label to make it "active"), the comment will not cause the thread to bump to the top. For example, this means that if a particular comment gets labeled as Offtopic, any replies "below" that comment will no longer bump the thread in the Activity sort. This will also apply retroactively, so if someone posts a new top-level comment, the thread will still initially bump to the top, but if that comment is then labeled as Noise, it will "un-bump" and return back to its previous location in the listing.
Since this will give us a better way to prevent threads from staying at the top of the site forever, I've also now changed the default time period back to "all time".
If you'd rather keep the previous behavior and continue having threads always bump to the top when a new comment is posted in them, you can use the new "All activity" sorting method instead. Logged-in users can set it as their default sorting across the site by changing to it on the home page and clicking "Set as default" to the right of the time period dropdown.
Any feedback is welcome, but these are questions that I'm particularly interested in:
- Are there cases where the label-based "uninteresting" judgment won't work well? Links to specific examples would be ideal, if possible.
- What other methods could we use to judge a new comment as "uninteresting"?
- Should we try triggering bumps from other non-comment events? For example, if a topic is getting voted up a lot, should it bump even if there isn't a new comment?
As usual, I've also given everyone 10 invites again (and don't worry, I haven't forgotten about turning the visible comment votes back on either, and I'll do that this afternoon, along with posting a thread to discuss it).
This looks like a really good "step one". Hopefully it encourages people to label things more; I certainly have not done so other than to try to use "exemplary" when possible.
I think finding alternate activity metrics are a really good idea, but other than votes, I don't really know what that would be.
One that I considered (and might try out at some point) is bumping the topic when a comment in it is labeled Exemplary.
My initial reaction is "that's a good idea" - I definitely think it's worth trying out.
Hmm... Interesting, to coin a phrase.
My first response is that it's confusing to have two sorting methods called "Activity" and "All Activity". I believe these names should be differentiated more, in order to make it clear that they're different under the hood.
The sorting method currently called "Activity" isn't really based on activity. The system now decides what types of comments will affect the sorting, rather than letting all comments affect this. And, you're asking if this sorting should be affect by non-activity events, such as voting and labelling. Effectively, you're going to make subjective decisions about how to present the threads in this sorting method.
I believe that the sorting method currently called "Activity" should be renamed to something which makes it clear that this is a complex sorting method which takes many factors into account when deciding how to sort the threads. I suggest something like "Curated" or "Relevance". You're curating the threads to present the more interesting ones first, or making sure that the more relevant threads are shown first.
The sorting method currently called "All Activity" would then be changed to "Activity", which makes it clear that this is a simple sorting method which merely responds to any activity in the threads.
I agree that it's pretty confusing to have them both with such similar names. I don't think I'd use "curated", since that implies a manual decision to me, like someone personally selected the posts. "Relevance" is okay but kind of overused in general, and I think only makes sense here if you say "relevant activity", which is quite long. Otherwise it sounds like it's saying something more like "topics the site thinks are relevant to you", which sounds more like a personalization algorithm.
Renaming "All activity" to something else might be reasonable too. Voting and labeling are "activities" too, and I don't think that "activity" necessarily means "comments only", even though that's how we've been using it here until now.
Is the intention that "All Activity" will always remain a simple sort based on the latest comments, with no other factors (while the complications get added to "Activity")? If so, if "All Activity" is just going to reflect the threads with the latest comments, that could become "Latest Discussion" or "Recent Discussion".
Or, if we renamed the existing "Comments" sorting method to something like "Most Traffic", we could then use "Latest Comments" or "Recent Comments" for the existing "All Activity".
I get that you want simple short names for the sorting methods, but that makes them ambiguous - especially when there are two comment-based sorting methods (one that sorts by most comments and one that sorts by latest comments).
Yeah, I think this is actually the biggest cause of confusion.
Some filters use thread information, others use comment information. The category names attempt to briefly cover both and that leads to confusion. eg. new comments vs new threads.
A tooltip explanation on hover may help some, but of course mobile users won't be able to see that.
Here's a crazy question. Is "Most comments" even a useful sort method? I understand sorting by new comments or threads, but simply sorting by volume seems unnecessary to me. Maybe the category name "Comments" can be reclaimed for something better.
Well, that's a whole different question. And a good question. I've been working on the assumption that we're going to keep all the existing sorting methods.
But it might be worthwhile to consider what sorting methods are actually needed. (@Deimos?)
If there is need (and I'm sure somebody will find a workflow that depends on it), then the less-useful sorts could be placed into a dropdown beside the others. Or possibly just remain accessible via URL, and pointed to in the docs.
Yeah, they don't necessarily always need to be easily accessible, but people do use that one reasonably often. It seems fairly common for people that don't visit frequently to do things like "most commented, last 7 days" to find the largest discussions from the last week.
Yeah, I actually made that worse as part of this change. The options were previously: Most votes, Most comments, Newest, and Activity, so having "Most" in there made it more clear.
Adding a 5th sort method meant that it took up more space and all the sort options would no longer fit next to each other on mobile with names that long, so I shortened some of them. I may have to use a different approach instead, like enabling the "tabs" to go wider than the screen on mobile and be able to be swiped back and forth, but that doesn't always work well.
Could you wrap the text in the names, so that they display across two lines?
“Relevant/General Activity” vs “Any Activity”?
I would like to avoid using the same word in multiple sorts, unless absolutely necessary. I think it would be more helpful for users if we keep the names as different as possible, to prevent confusion.
"Discussion" vs "Noise"?
Using "Noise" as the name of a sorting option makes it looks like this option is going to sort the noisiest, most off-topic, threads to the top.
I want more noise in my Tildes! :)
Does it solve the problem of two users having a conversation for days on end? I don't know how often labels are used in practice, but I expect the majority of comments go unlabeled.
Perhaps if a chain consists entirely of two users, that should be deemed not interesting.
Not yet, unless someone labels that thread (which isn't always appropriate).
It could definitely be updated to cover it though, any thoughts about specifically how to detect it? For example, maybe if a chain is at least 4 comments replying to each other where only 2 users are involved? That would be two replies each back and forth.
That seems like a fair approach to me. After four or five replies threads usually turn off topic or devolve into pedancy ("no, you said XYZ above").
Deeply-nested, three-way arguments are a lot less common in my experience so scaling it up to more users probably isn't required.
Out of curiosity, does the bumping logic occur when a comment is actually posted/labeled (eg. updates an activity score in the DB)? Or is it calculated on-the-fly when building the list of threads for the reader?
If the latter, I imagine it would get expensive to start applying more complex logic when building an index like this. Is that right?
Ninja edit: Actually I guess it's not generated for the reader specifically because there's no user personalization. So it should be a lot cheaper to generate, even if it is all dynamic.
It's much closer to the former: there's a background process that gets triggered whenever a "potentially interesting" event happens on a topic, which right now is when a new comment is posted in it, a comment in the topic is deleted or removed, or a comment in the topic gets a label added or removed from it. That process looks at the thread and decides when the "last interesting event" happened, and stores that time on the topic. Then the activity sort just uses those times.
Gotchya. So that means that adding some complex logic might affect the parsing time of the background process, but has utterly no effect on generating the link index. Thanks!
I would suggest not taking comments above a certain nesting level into account. This is not perfect, but if the threshold is level 10 then long conversations, even if on-topic, will not bump a topic.
Maybe have the option to also self-report not to bump the thread that is only available to comment replies? By default, a user would have to opt-in to having their comment not bump the thread.
The ability to self-label a comment might be nice. Give a user access to only Off-Topic, Joke, or Noise, give it full power, and you have a 'sage' function that's opt-in.
This would be covered by the so-called "whisper comment" feature which keeps getting requested.
The most recent comment on that issue is from Deimos, two months ago:
With this sorting change announcement, it sounds to me as though the groundwork's been laid to proceed with implementing this (perhaps called "asides," which I prefer).
It should be possible to disable this feature (the back and forth) if it is implemented. I am personally interested in whatever activity unless labelled explicitly.
Another potential feature could be a finer grainer control of labels: I don't mind off-topic activity per se, but I'd rather filter out noise or malice or jokes.
I assume that would be available via the "All Activity" sorting method, which doesn't make any judgements about which comments should bump a topic: all comments will bump a topic in "All Activity".
To make it more concrete: I would like to have comments tagged [off-topic] to bump threads up for me, and maybe even [joke], but not [noise] or [malice]. "All activity" means that anything bumps it up, both the back and forth, and the (personally) undesired comments.
It's not feasible to maintain custom sorting orders for individual users.
Sure, but can't the algorithm be parameterised by options in the settings page? E.g.:
He means, I believe, that for each of the sorting methods, there is a pre-computed ordered list calculated once then served to everyone. Having parameterized method like you suggested wouldn’t be hard to implement, but it would mean maintaining
n!n2 additional precomputed ordered lists to service any combination of parameters someone might choose.It would be n2 since each filter is an included/not included binary and they aren't order sensitive.
Good catch!
Isn't this the case with subscriptions already, tho? Not everybody is subscribed to everything.
Calculating sorting weights for each topic is a lot more comparatively costly, so gets done once for each. Simple filtering of that already sorted list by tags once is cheap and probably done on the fly.
or an ancestor limit on a comment maybe? Like if the comment has four or more ancestors, don't bump it? [just a rough idea]
Oooh I like this. At least conceptually I do, we'll have to see how it goes.
A minor concern I have is the name, and I realize that's nitpicky especially if I don't have a perfect alternative. "Interesting" to me seems like a value judgement of the content being shown to the user looking and reminds me of algorithmic recommendations based on viewing history or whatever. I know that's not what it is, but imo it's not a great look. I almost wonder if calling it anything other than "activity" would even be necessary if it became permanent, I don't think taking out unsubstantial activity violates what "activity" implies too much. But maybe someone has a genius alternate title, anything I can think of is too wordy or vague.
I also wonder if "Offtopic" should maybe be exempt from this. While it's hard to see much of a reason why comments that deserve the other labels would be something others would want to see / that we would want to encourage others seeing, an off topic conversation has plenty of room to be valuable and "interesting," it just shouldn't be allowed to dominate at the top of a thread as it's starting. The effect of the label has been pretty minimal up until now, but now it is a huge deal and I'm not sure that's justified.
Nothing on the site actually says "interesting", that's mostly just me trying to find a decent way to describe the goal of it (and what it's referred to internally, but almost nobody will see that). It does just call it "activity" on the site, and "relevant activity" in one location (the dropdown on a user page's Topics page).
I'm not certain about totally disregarding the Offtopic threads either, but we do seem to often have posts where there's some ongoing argument about something that's only tangentially related, and that's what keeps bumping the thread up forever.
Ahh okay, didn't realize until now that it was live.
I wonder if we should try making it so we can identify the difference between an off topic conversation that is fine and a long chain of bickering. Maybe a label for drawn out back-and-forths which only impacts this sort? I guess it could be argued that those conversations eventually become noise but that isn't really how we're using that label as of now.
Agreed, I'm not sure what the best way to handle that is, but I don't think anything in the current set of labels does it.
I've thought a bit about a label like "argumentative" that could work to exclude it from this sort, and maybe even have some other effects like slowing down how quickly people can reply, to prevent those quickly-escalating cases when people reply impulsively.
I was thinking about this same thing with the word "contentious." I've hardly ever used the "malice" label because, to me, it feels like it has a pretty high threshhold. I feel like you have to show an outright intent to harm someone else.
What I see more often, however, are people who get into a negative back-and-forth, and while neither of them are making threats, the tone of the conversation is tense, they've long forgotten the principle of charity, and everything is sitting precariously on the edge of explosive hostility. Being able to tag this in a way that doesn't convey "needs removal" but conveys "warning" or "slow down" could be very beneficial. I know I've recently been involved in some emotionally charged threads, and I had to deliberately give myself time away from the screen in order to respond in a way befitting the goals of the site and not one that would just let me take quick jabs at others. Building that kind of cool-down time into the site could benefit tempers and encourage self-reflection.
Maybe make it so that comments at different depths don't count unless there is already a certain amount of replies to the parent comment? For an aggressive example, each top-level-comment counts, but second-level-comments only bump if the top-level-comment already has a reply. This way back & forth conversations don't cause any bumps after the first comment, but if a comment sparks many replies, it is counted. A similar method would be to count every second, third, etc. comment for a given depth.
Another simple rule could be to always count comments if the user making it has no other comments in that thread.
HackerNews has a timer for nested replies. The deeper a thread goes, the longer it can take for the 'reply' button to appear. This method seems to prevent threads from going too deep during arguments, but maybe also discourages discussion.
Perhaps a system by which the activity value of a nested comment decreases the further down the reply chain it goes would be an effective strategy: A new top level comment is interesting because it starts a new train of thought and therefore bumps the post to the top of the activity list, but a really really nested comment might only be interesting to the people that are part of the discussion, so the post gets bumped up by a less significant amount.
Is that how it works? Been on the site 8+ years and had no idea. I usually just click the permalink for the comment I want to respond to when there's no reply button.
I don't have any evidence for this other than hearsay. I usually don't comment on HN threads so the lack of a reply button is never a problem for me.
Hacker news has a whole slew of undocumented features, which is one of the worst parts of the community IMHO.
As I've said elsewhere, finer grain control of what is interesting would be a nice feature to have. Tho there is the fact that off-topic threads are not marked off-topic if they are inherently interesting enough, so that might be unnecessary.
I comment toplevel to reply to this:
I think this is a nice idea. We could have a system where every 5-10 votes has the same affect of an "interesting" comment, or even one where after each bump the votes necessary is increased (e.g. 1st bump after 5 votes, 2nd after 6, 3rd after 7, and so on), given both highly voted stuff and stuff that appears higher up and stays there tends to accumulate way more votes than the rest.
Since this has been around for a while, I thought it would be worth giving some feedback...and all I can really say on that front is that it's basically identical. I've been switching between the views every once in a while to make sure, and I've never seen them be different. That's not to say that they haven't been, but it's never been something I could detect even while going out of my way to compare them. So...whatever problems this was meant to solve, I don't think it has...but it also hasn't made anything worse, so that's good!
Is this just my experience, or has it been pretty much business as usual for everyone else?
There haven't been many effects, but I have noticed it prevent (or reverse) bumps due to Noise/Offtopic comments a decent number of times.
It's also the nature of it that you can't really tell how much it's doing, because there's no way to know when something doesn't happen because of it. For example, if someone posts under a Noise comment, previously that comment might have set off a discussion or argument, which would have made the thread keep bumping back up, drawing more attention and continuing it even longer. Now, since it won't bump for the large majority of users, maybe that ensuing discussion just never happens instead. It looks exactly the same as no effect, but it was actually a significant one.
Overall, I think it probably hasn't been too impactful so far, but I'd definitely also like to start adding in some of the "additional" bumping methods, for things like reaching vote thresholds, a comment getting Exemplary-labeled, and so on. That might have more of an effect, and then it'll be even harder to compare directly since it's coming from activity types that the old Activity sort doesn't even consider.
my immediate problem with this (and why in its current form i'll likely not use it at all) is that this will almost certainly punish threads (and honestly potentially kill some of them) in the event that they get a reply, but that reply happens to be construed as bad by people. i get that we're trying to prevent certain threads from dominating everything and trying to disincentivize certain types of replies, but how this works is inelegant to me, to say the least, especially given that many interesting topics already get posted to this site to no discussion because we have more content submitted daily than people who can hold discussions on that content. i feel like there are ways to do this that don't have the significant side-effect of seemingly throwing some good content under the bus with the bad.
I'm not sure how this is punishment for those threads. Near as I can tell it's not that they're being artificially pushed down but that if that single reply gains a label, it reverts to as if it never received a comment at all. It doesn't award them if the first comment is "lol tru," but that seems far from punishment.
Unless you think noisy comments are unusually common in small threads and the removal of the "lol tru"s of the website will impact them more, but I kinda doubt that's the case.
It's possible, but as always, if I see people misusing labels I reduce their "label weight", which means they have less influence to be able to make labels "active".
This also isn't the only sorting method (and it's a method that some other sites don't even have), and lots of people use other ones like sorting by votes, newest topics, etc. that aren't affected by it at all. We'll have to see how it works out in practice, but if a thread gets entirely killed by a comment getting marked as Noise then it probably wasn't actually attracting much interest anyway.
well, my thing is mainly that that topics further down even the first page pretty much die beyond a certain point right now. even though tildes isn't against bumping of older threads, very few people seem to reply to threads once they get past about 3/4ths of the way down the first page, and threads past page 2 almost always die and never get replied to again, so even a small change in how activity is prioritized and sorted right now can lead to threads that might have been viable getting fucked over unintentionally by a change like this. it's one of those weird things that comes with our current level of activity, and i'm not sure that it'd even be a big problem if we were slightly less active or more active than we currently are.
But that seems like even more reason not to bump threads whose only new replies are offtopic, noise, etc., because bumping them pushes other posts down the page.
no? because the problem is the reply, not the actual thread. it doesn't make sense to basically punish a thread because its singular reply is shit, because a single reply is a single reply and not representative of the other possible replies to a given thread.
If there are other replies that aren't noise/offtopic/malice those replies will bump the thread. The only thing this does is disregard the noise/offtopic/malice replies.
my entire point is that there are not other replies in my hypothetical example.
If there aren't any new replies that are on topic why should the post be bumped?
Maybe the answer to that is to slow down a little bit on posting content so that people can keep up. ;)
that's not really something that is solved by arbitrarily decreasing the rate of posts on here. (if anything, that would probably exacerbate the problem because it's a lot more multifaceted than you're making it out to be here.)
Yeah, this is still something that I'm planning to do eventually. At least now if someone labels your comment Noise it will un-bump the thread, so this gets a little closer.
It's probably not the way I'd like it to work overall, but it might be possible as a temporary method to allow users to label their own comments with some of the labels (maybe Noise, Offtopic, Joke).
Setting any labels before comment submission would negate the need for a time delay.