That's one featurebox myth busted · 7:19pm Jun 6th, 2014
It's a commonly discussed complaint — one I actually believed myself — that there's something in the coding for the final three slots of the featurebox that explicitly privileges older stories; I've heard multiple times that it seems impossible to find a story newer than mid-2013 in it. If so, either that code has changed, or folks trying to figure out the mysteries of the Featurebox Update Slots are barking up the wrong tree.
Never The Final Word, my open anthology of continuations of other authors' stories, was first published a month ago (2014/04/30). It never came close to the featurebox on original publication — anthologies and short-story collections rarely do; they get very little attention in the grand scheme of things. That's why I was so surprised when I glanced at the front page a few minutes after updating it with two new submissions, to find it sitting prettily atop the update slots.
(The other two featured stories shoot the idea down even harder. Conversations in a Canterlot Café was published a mere 13 days ago and just posted its third chapter. Dreams About Friendship Are Magic was published on 2014/03/28, and just published its 11th chapter. Not a single one of the features is over three months old.)
So what is going on with those update slots?
It seems logical to me that the bottom three slots are based on rating (the same upvote- and downvote-based calculation that drives the Top Stories list) rather than "heat" (what drives the top 7 slots of the featurebox — it takes into account views, upvotes, and comments per unit of time, which heavily favors just-published stories). The whole point of those bottom three slots is to spotlight continuing stories whose heat has long since cooled off. The current featurees, at first glance, support this — respectively, the three update slots have votes of +61/-0, +218/-2, and +251/-8.
This idea gives us some data to play with — since the relative ratings of an author's stories can be viewed in the sidebar alongside one of their blog posts or story pages. (If you're reading this in your feed, click the title of my post or click here, and you'll be able to see mine over to the right.) That extra information quickly complicates the situation.
Because of the weird way rating is calculated [1], a 251/8 ratio is worse than it looks. In my own sidebar, No Regrets (+264/-8, almost identical to Dreams About Friendship Are Magic) is languishing down in 6th place, with Hard Reset 2: Reset Harder (+853/-27) three spaces above it. All other things being equal, if DAFAM hit the update slots, HR2 should fairly consistently be doing so. But HR2, which featureboxed high and early, AFAIK hasn't shown up in the update slots since.
Also, because of the weird way rating is calculated [2], No Regrets' +264/-8 score makes it higher-rated than Never The Final Word (+61/-0), which is in seventh place in my sidebar. So if it went strictly by rating, DAFAM (+251/-8) should have been higher in the 'box than NTFW. There's another factor at play, which I don't have a good hypothesis for yet.
One other note. I've heard the suggestion that once a story has featureboxed, it's not eligible to get into the update slots a second time. I don't have any definitive data points for that, except for HR2's failure to 'box. The best evidence here would be a story which has featureboxed already showing up in the box again (disproof), or to track stories which have featureboxed already — and ideally haven't accumulated any new downvotes — and confirm that they update without showing up.
A variation of this is that there's a "cool-off" period during which a previously featured story is ineligible to 'box again, to prevent short updates from spamming a story repeatedly into featuring. This is plausible, since the Updated Stories box on the main page already works like that — a story can only be listed in it once per day. The featurebox cool-off might be longer, perhaps a lot longer. If, let's say, a story can only featurebox once every six months, that could explain the observation about old stories disproportionately showing up. I might be able to offer a test case for that by updating Princess Luna Picks Up Hitchhikers, my second-highest-rated story, which last updated (and 'boxed in the update slots) almost a year ago.
Any thoughts? Theories? Speculate away in comments.
--
[1] Downvotes, especially the first few downvotes, have an amazingly outsized effect. Based on recent observations (such as the rise and fall of arcum42's Cubic Zirconia from the Top Stories list), the difference between +200/-0 and +200/-1 is enough to take a story from the very top of the Top Stories list to off the first page, and +200/-2 makes it vanish from the list entirely.
Edited to add: Since this blog post's publication, Never The Final Word got 3 more upvotes — and then its first downvote. +64/-0: Sixth place. +61/-0: Seventh place. +64/-1: Ninth place. See what I mean about that outsized effect?
[2] Also, rating seems to take into consideration the confidence interval of the upvote/downvote ratio: +2000/-10 scores far higher than +200/-1, even though the ratios are identical. (There's actual math behind this — the idea is that, if you take a +200/-1 story and sic 1800 readers on it, the odds that you will end up with more than 10 downvotes are substantial, so you can be more confident that the one which already has 2000 votes reflects a genuine 200:1 ratio. We've got a lot of statisticians here, though, so I'll let someone in the audience go into more detail.)
I'm pretty sure there was a code change fairly recently. Cheerilee's Thousand is almost always featured there when it updates now, but it didn't start that until a month or two ago.
It's been explained to me that the three updates with the strongest heat hit those slots now, regards of when they were originally published.
I've also been told that a story can only be featured within eight hours of it's initial publishing. It is ineligible after that. I'm not sure if that is true or not.
I think the third factor in ratings calculations is a comparison of the number of ratings to the number of views that generated them, so getting fewer votes on the same number of views results in a lower-rated story, and getting more votes on the same number of views results in a higher-rated story.
I believe this is the case because the high-rated story list seems to have a bias against stories that are old enough to predate the full 'establishment' of FimFiction. If you got on EqD back when lots of people still hadn't made accounts here, you would end up with a lot of views without so many favorites/tracks and upvotes, and stories where this is the case seem to suffer.
No need to drive yourself nuts. There was a bug in the system that prevented anything after 2013ish to be featured in the last three slots. Obs got on the programmer's ass (can't remember which one) and it was fixed.
Feature away!
I've seen a couple of mine resurface in the bottom three slots for an hour or two when I update, but nothing dramatic.
2181136 2181196
Thank you for sharing your knowledge!
It's learning things like that which make this navel-gazing time worthwhile.
2181138
Interesting thought. I'm not sure how to separate views out as an independent variable, unless there are sets of stories very similar in votes but dissimilar in views that you can sort by rating and see if they go out of expected order.
Re a bias against early stories, "Sunny Skies All Day Long" is solidly in the top stories list — with the caveat that it was first published in early 2011, but the FIMFiction version shows a date of 2012. I think a lot of those earlier stories later migrated here from google docs, which could contribute to a skewing of statistics.
Has "Never the Final Word" been featured before? I was under the impression that the last three slots go to stories that have been featured before when they update. So when you put up the next chapter of "Reset Harder," for example (I'm pretty sure it's been featured already, right?), I bet it will show up in one of those bottom three slots, but won't go any higher than that.
2181446 I think Sunny Skies got a nostalgia bump when it got added here, a bit like Dangerous Business got recently. SPark and I came up with this idea when looking at some of her earliest stories, which were never published on gDocs.
I can't think of a way to untangle votes/views as an independent variable either. It would make a good confounder, though--with votes/views, vote ratio, and vote ratio certainty weighting you've got enough variables in play that it makes it very difficult to reverse-engineer, which is desirable when you don't want people to be able to play the system.
It's update time. The most recent update is on top of the three slots and it is pushed down by new updates until the third high-rated update pushes it out of the tenth feature box slot.
2181460
The system does not, as far as I know, even keep track of which stories have been featured. I think it relies on some sort of system for picking out which recently updated story to feature, I strongly suspect based on some rating threshold.
2181446
I will note that the site RATING system I have determined to be wholly based on upvotes and downvotes and absolutely nothing else whatsoever. HOW it is calculated, I cannot tell you, but stories with identical upvote:downvote ratios are always at the same point in the rating scale. I will note that the more upvotes you have, the less downvotes count against you; a story with 900 upvotes and 9 downvotes is considerably better off than a story with 100 upvotes and 1 downvote. The larger the absolute number of upvotes, the worse your ratio can be.
I have not disentangled the heat system yet, but heat and feature box placement are not identical. I have heard that heat is based on upvotes per unit time, comments, and favorites; I'm not sure if that's true or not. Upvotes is definitely the case; the other two I am less certain of. I am fairly certain that downvotes either do not hurt a story's heat, or do not hurt a story's heat very much.
2181921
One other note:
The highest rated stories are actually packed INCREDIBLY closely together. Going from 900:9 upvotes:downvotes to 911:9 upvotes:downvotes took a story from #43 to #27. While the very front page may not be quite as tightly packed, it is still pretty close. A single downvote on any story will send it downwards. And the smaller the number of votes a story has, the more each vote counts for (not surprising, I know, but still important when you're considering how the system works).
I can tell you that going from 330:0 to 330:1 left a story on the very top of the rating list, though, so while downvotes do have an outsize effect, 330:1 is still "tops". 330:2 was way down the list, though.
With all due respect, I think you're all overthinking this. I can verify personally that there's no proscription against "new stories" topping out in the feature box, at least not as of a few months ago. My first real story hit #2 in the feature box last summer, within 24 hours of being published. The sequel, six months later (November), made #1 and stayed there an entire weekend. Before that, I had like 8 followers, so this was a total cold start.
That said, it's clear the box is divided. The last three slots in the box are clearly are reserved for updating stories. The top seven are reserved for new stories only. Thus, there's basically two boxes, new (top 7), and updates (remaining 3.)
Overall, I think it IS a pure heat rating beyond that. Stories at the top of the "popular stories" list push directly up into the bottom slot of their respective (new or update) category. Likewise, stories that drop off the bottom end up at the top of that list. At least when I was last paying serious attention a couple of months ago. (I'm a horrible updater lately.)
"Heat" is simply a decay-based metric, with things like favorites, likes, views, etc. given varying weights, and a story's total weight decaying over time. In fact, I think downvotes may actually be a _slight_ positive in the super-short term because of this, but that's highly speculative. I don't know the exact weights of course, nor the decay/fall-off rate, but it's almost certainly exponential from what I can tell. This is a very common system for social media rankings all across the internet, and as far as I can tell, looks to be exactly what fimfic uses. Nothing here seems unusual compared to these known and well described systems. The only unknowns (per my opinion) are the exact weightings and decay rates used. And why would it? Knighty doesn't need to reinvent the wheel, just tune an existing one for our community.
2183043
Right. We're just talking about the "update" slots, though (the final three). At least as of a couple of months ago, there was pretty widespread talk that stories published after mid-2013 would not reach the featurebox's update slots when they posted new chapters, no matter how popular they were.
As 2181136 and 2181196 said, this seems to have in fact been the case, but was recently fixed. Yesterday was the first time I'd realized things were different.
2184676 I are idiot.
Somehow I misinterpreted "final three slots" as the top 3... like the final four are the four best teams. My bad!
The Great Alicorn Hunt seems to hit the feature box just about every time it updates, so it's clear that a story can hit the feature box more than once... though there could still, of course, be a cooldown period before it's allowed to do so again.
tl;dr, but I know HOW the features box works, I've talked to the admins and moderators first hand.
I don't know if your knowledge has changed since this post, but here's how it works.
The bottom three slots are exclusively for updated stories. When a popular story gets updated, chances are it'll get featured within 5-10 minuted of updating. I've seen it happen with Researcher Twilight and Diaries of a Madman.
That said, the other seven slots are for stories that have been recently published within the last day or two. The more popular the story is, the higher up on the list it will get.
It's not just upvotes/downvotes or whatever point you were trying to make, the features box is designed with an algorithim that takes into account upvotes AND downvotes, how long it's been published, favourites, and so on. Summarising, the more attention it gets within a shorter period of time the more likely the chance it will get featured, that goes for new stories and updates ones.
On a side note, I've had two stories get on the features box in recent times from publishing off the bat, woken up to find my notifications explode harder than Nagasaki. That, and because I was updating my first story when I joined this site on a daily basis, it got one of the bottom three slots on the box at one point. I wasn't awake to see it, but one of my friends told me this was the case.
One more thing; IT IS ENTIRELY AUTOMATED, the site staff have nothing to do with it. The popular stories work in the same manner, but those ones are more concerned with the 'heat' surrounding them, for lack of a better word.
I don't know where your idea of the ratio of likes to dislikes came from, but that's only a small fraction of it. One story I wrote got features and at the time my ratio was two likes for every dislike, so it had a ratio of 2/1, making a third of its total count dislikes (It was a fetishised rape story, and people were still enjoying it
).