What do favorites look like? December 2, 2011 9:12 AM Subscribe
What does the slope of favorites look like for FPP or AskMe questions?
That is, taking a random post with a large enough amount of comments for a sample (100+ maybe?), and take just the comments with favorites. What does the drop in favorites look like as interest in the FPP/AskMe wanes? Does it look any different when there is a very large amount of comments (in the 500-1000+ range)?
That is, taking a random post with a large enough amount of comments for a sample (100+ maybe?), and take just the comments with favorites. What does the drop in favorites look like as interest in the FPP/AskMe wanes? Does it look any different when there is a very large amount of comments (in the 500-1000+ range)?
This is totally calculable from the Infodump, if anyone feels like having a crack at the favorites data. I'd guess on average you'd see a pretty quick and significant tailing off after the first few hours but with a lot of noise in individual threads; how the busyness of the thread would effect that is an interesting question.
posted by cortex (staff) at 9:30 AM on December 2, 2011
posted by cortex (staff) at 9:30 AM on December 2, 2011
Are you thinking about favoriting rate in the thread per unit time, or the number of favorites a comment gets based on its position in the thread?
posted by auto-correct at 9:31 AM on December 2, 2011
posted by auto-correct at 9:31 AM on December 2, 2011
The latter would be a function of the former, right?
posted by griphus at 9:32 AM on December 2, 2011
posted by griphus at 9:32 AM on December 2, 2011
Depends on your terms. Comment #10 and comment #100 in two different threads could have the same temporal position in their respective threads, and the open question there is whether the rate of commenting in a thread would have some correlation to differing rates of favoriting over time.
posted by cortex (staff) at 9:37 AM on December 2, 2011
posted by cortex (staff) at 9:37 AM on December 2, 2011
So I guess you'd want to maybe look at buckets of threads with different rate-of-commenting profiles over, say, the first 24 hours of comments in the thread and see if the average profiles of those different buckets differ in interesting ways.
posted by cortex (staff) at 9:38 AM on December 2, 2011
posted by cortex (staff) at 9:38 AM on December 2, 2011
I guess the original question is whether there's a global profile, or if there are a number of individual profiles depending on comments/commenting rate/etc.
posted by griphus at 9:40 AM on December 2, 2011
posted by griphus at 9:40 AM on December 2, 2011
Partly, but I think they're two different questions. Like, how early questions accumulate favorites over time vs late questions gives you different information than what the Graphfi histogram gives you. I'd been meaning to look at the former recently, maybe this will get me going.
posted by auto-correct at 9:40 AM on December 2, 2011
posted by auto-correct at 9:40 AM on December 2, 2011
A global profile to favoriting, that is, not to commenting.
posted by griphus at 9:40 AM on December 2, 2011
posted by griphus at 9:40 AM on December 2, 2011
(that was in response to your earlier comment)
posted by auto-correct at 9:41 AM on December 2, 2011
posted by auto-correct at 9:41 AM on December 2, 2011
Ugh, overt racism? Really?
This smacks of privilege.
posted by Eideteker at 10:25 AM on December 2, 2011 [1 favorite]
This smacks of privilege.
posted by Eideteker at 10:25 AM on December 2, 2011 [1 favorite]
Eideteker rejects your Euro-centric thoughts on graphing, apparently.
posted by Think_Long at 10:31 AM on December 2, 2011
posted by Think_Long at 10:31 AM on December 2, 2011
This invisible graph paper is really hard to work with.
posted by griphus at 10:32 AM on December 2, 2011 [1 favorite]
posted by griphus at 10:32 AM on December 2, 2011 [1 favorite]
What do favorites look like?
Some of mine are ugly and misshapen, like ashtrays made at pre-school, but I love them anyway. I even bring them out once a year around the holidays to show the relatives, look at the adorable favorite I got!
I've got my first favorite framed and hanging on the wall of my rumpus room next to my Nagel print.
posted by Ad hominem at 10:34 AM on December 2, 2011 [1 favorite]
Some of mine are ugly and misshapen, like ashtrays made at pre-school, but I love them anyway. I even bring them out once a year around the holidays to show the relatives, look at the adorable favorite I got!
I've got my first favorite framed and hanging on the wall of my rumpus room next to my Nagel print.
posted by Ad hominem at 10:34 AM on December 2, 2011 [1 favorite]
What does the slope of favorites look like for FPP or AskMe questions?
Methinks it is like a weasel.
posted by Wolfdog at 10:35 AM on December 2, 2011 [1 favorite]
Methinks it is like a weasel.
posted by Wolfdog at 10:35 AM on December 2, 2011 [1 favorite]
...next to my Nagel print.
You joke but I had to be talked out of buying one recently.
posted by griphus at 10:38 AM on December 2, 2011
You joke but I had to be talked out of buying one recently.
posted by griphus at 10:38 AM on December 2, 2011
You really need a Testarossa if you are going to buy one. I'm saving up till I can get both at the same time, don't want to be one of those guys with a Ferrari and no Nagel print.
I've thrown away all my socks in preparation.
posted by Ad hominem at 10:45 AM on December 2, 2011
I've thrown away all my socks in preparation.
posted by Ad hominem at 10:45 AM on December 2, 2011
I get orange from the f and a strong purple from the v, nothing else aside from some red at the end on e and a vague yellow coming from I dont know where. I guess favorites look like a sunset to me. Or Barney being set on fire.
posted by villanelles at dawn at 11:06 AM on December 2, 2011
posted by villanelles at dawn at 11:06 AM on December 2, 2011
Ugh, overt racism? Really?
This smacks of privilege.
posted by Eideteker at 6:25 PM on December 2 [
Wut?
posted by Decani at 11:36 AM on December 2, 2011
This smacks of privilege.
posted by Eideteker at 6:25 PM on December 2 [
Wut?
posted by Decani at 11:36 AM on December 2, 2011
Presumably Eideteker would have preferred the lede to be "What does the person of Asian descent of favorites look like for FPP or AskMe questions?".
posted by cortex (staff) at 11:39 AM on December 2, 2011
posted by cortex (staff) at 11:39 AM on December 2, 2011
griphus I have a book of Nagel prints ( don't ask) if you want them.
posted by The Whelk at 12:03 PM on December 2, 2011
posted by The Whelk at 12:03 PM on December 2, 2011
What does the slope of favorites look like for FPP or AskMe questions?
slippery.
posted by quonsar II: smock fishpants and the temple of foon at 12:04 PM on December 2, 2011 [3 favorites]
slippery.
posted by quonsar II: smock fishpants and the temple of foon at 12:04 PM on December 2, 2011 [3 favorites]
I don't think this answers the question as asked, but maybe it provides a starting point for further exploration: Favorited comments over time for 100 Metafilter posts.
Based on the Nov 26 infodump. Each box represents a Mefi FPP; posts 98000 through 98099 are shown. The x-axis of each box is time, 30 minutes per pixel, 24 hours shown in total, starting from the time of the post. The y-axis is the number of favorites made on comments on that post in each 30-minute interval, one pixel per favorite (to a maximum of 48, which was exceeded in a handful of cases).
posted by stebulus at 2:13 PM on December 2, 2011 [2 favorites]
Based on the Nov 26 infodump. Each box represents a Mefi FPP; posts 98000 through 98099 are shown. The x-axis of each box is time, 30 minutes per pixel, 24 hours shown in total, starting from the time of the post. The y-axis is the number of favorites made on comments on that post in each 30-minute interval, one pixel per favorite (to a maximum of 48, which was exceeded in a handful of cases).
posted by stebulus at 2:13 PM on December 2, 2011 [2 favorites]
This version is crisper, since it has the lines correctly aligned to the pixels.
posted by stebulus at 2:47 PM on December 2, 2011 [3 favorites]
posted by stebulus at 2:47 PM on December 2, 2011 [3 favorites]
Stebulus' graphic is interesting - many FPPs show 2 peaks of favorites, the first peak in the first 5 minutes or so, the second after the post has been up about 15 minutes. I'm guessing the first peak may be a lot of "looks interesting; I'll mark this for reading later" and the second peak comes after people have actually read the link.
It would be interesting to see if there's a correlation between time of faving and commenting behaviors - under this hypothesis early favers (who haven't read the link yet) would be more likely to post drive-by snark and one-liners than those who RTFA and faved afterward. Panoptical, but for SCIENCE!
posted by Quietgal at 2:48 PM on December 2, 2011
It would be interesting to see if there's a correlation between time of faving and commenting behaviors - under this hypothesis early favers (who haven't read the link yet) would be more likely to post drive-by snark and one-liners than those who RTFA and faved afterward. Panoptical, but for SCIENCE!
posted by Quietgal at 2:48 PM on December 2, 2011
And here's my code, in the interests of transparency and such. (Criticism welcome.)
the first peak in the first 5 minutes or so, the second after the post has been up about 15 minutes
The time resolution of the graphs is 30 minutes.
"looks interesting; I'll mark this for reading later"
Just to be clear: The graphs show only favorites on comments, not favorites on posts.
posted by stebulus at 2:54 PM on December 2, 2011 [1 favorite]
the first peak in the first 5 minutes or so, the second after the post has been up about 15 minutes
The time resolution of the graphs is 30 minutes.
"looks interesting; I'll mark this for reading later"
Just to be clear: The graphs show only favorites on comments, not favorites on posts.
posted by stebulus at 2:54 PM on December 2, 2011 [1 favorite]
Ah crap, I thought it was faves on FPPs. Another beautiful hypothesis bites the dust. (And I thought the whole box spanned 30 minutes, not that each bar was 30 minutes. So much for my reading comprehension.)
posted by Quietgal at 3:00 PM on December 2, 2011
posted by Quietgal at 3:00 PM on December 2, 2011
I get orange from the f and a strong purple from the v, nothing else aside from some red at the end on e and a vague yellow coming from I dont know where. I guess favorites look like a sunset to me. Or Barney being set on fire.
villanelles, are you a synaesthete, you bastard?
posted by Diablevert at 3:04 PM on December 2, 2011
villanelles, are you a synaesthete, you bastard?
posted by Diablevert at 3:04 PM on December 2, 2011
Just enough to wish I was much more. I see colors especially strongly in numbers, but can't do any of those neat tricks like instantly pick out a few scattered 2's in a field of 5's.
posted by villanelles at dawn at 3:12 PM on December 2, 2011
posted by villanelles at dawn at 3:12 PM on December 2, 2011
many FPPs show 2 peaks of favorites
I think time of day might be a factor. The majority of MeFites live in North America. The site becomes less active when North Americans are asleep. If a post is made late in the day in NA, the thread can pick up a second flurry of posting activity the following morning, and presumably have a delayed or flattened peak if it's made late at night.
FPP 98012, one of the ones that really sticks out as double peaked, was made at 8:09 PM and contains this comment.
posted by nangar at 5:02 PM on December 2, 2011
I think time of day might be a factor. The majority of MeFites live in North America. The site becomes less active when North Americans are asleep. If a post is made late in the day in NA, the thread can pick up a second flurry of posting activity the following morning, and presumably have a delayed or flattened peak if it's made late at night.
FPP 98012, one of the ones that really sticks out as double peaked, was made at 8:09 PM and contains this comment.
posted by nangar at 5:02 PM on December 2, 2011
We looked at this type of thing in various different ways back in December 2009. It sort of starts with this comment from me and I kind of ramble on a bit over the next little while and post a bunch of links to various charts.
I think where we ended up was with this comment that links to a chart showing average favorite count on comments vs. their position in the thread (as a percentage of the total thread length). That turned out to be the best way to compare threads of different lengths and see if comments near the beginning of a thread tend to get more favorites than those near the end.
posted by FishBike at 5:22 PM on December 2, 2011
I think where we ended up was with this comment that links to a chart showing average favorite count on comments vs. their position in the thread (as a percentage of the total thread length). That turned out to be the best way to compare threads of different lengths and see if comments near the beginning of a thread tend to get more favorites than those near the end.
posted by FishBike at 5:22 PM on December 2, 2011
With Ask my favorites always drop off substantially once the question is off the front page, and it's unusual for many favorites to be acquired if I didn't make the answer in the first hour or two after the question was posed. My one shining example of a well-favorited answer got two random favorites a month and a half later, which I don't really understand.
And it might just be my impression, but 2nd or 3rd in the thread (not 1st) appears to be a magical favorite-gathering position.
(I bestow favorites freely without regard to when they were posted, and since I lurked for years and years I often find myself hunting down old posts and favoriting away without noticing how out of date it all is. I'm that wacky favoritor who hit the plus sign on that post from 2005 that you totally forgot you made. I also may secretly have a need to mess up pretty and logical histograms - I don't think there's a support group for that.)
posted by SMPA at 8:07 PM on December 2, 2011
And it might just be my impression, but 2nd or 3rd in the thread (not 1st) appears to be a magical favorite-gathering position.
(I bestow favorites freely without regard to when they were posted, and since I lurked for years and years I often find myself hunting down old posts and favoriting away without noticing how out of date it all is. I'm that wacky favoritor who hit the plus sign on that post from 2005 that you totally forgot you made. I also may secretly have a need to mess up pretty and logical histograms - I don't think there's a support group for that.)
posted by SMPA at 8:07 PM on December 2, 2011
Argyle socks are the bomb.
posted by flabdablet at 6:06 AM on December 3, 2011
posted by flabdablet at 6:06 AM on December 3, 2011
nangar: I think time of day might be a factor.
I didn't notice anything addressing this exact point in FishBike's impressive collection of graphs, so I thought I'd take a stab at it:
The site becomes less active when North Americans are asleep.
Favorited comments on Metafilter by time of day.
If a post is made late in the day in NA, the thread can pick up a second flurry of posting activity the following morning, and presumably have a delayed or flattened peak if it's made late at night.
Favorited comments over time for 10,000 Metafilter posts, grouped by post's time of day.
posted by stebulus at 10:16 AM on December 4, 2011 [1 favorite]
I didn't notice anything addressing this exact point in FishBike's impressive collection of graphs, so I thought I'd take a stab at it:
The site becomes less active when North Americans are asleep.
Favorited comments on Metafilter by time of day.
If a post is made late in the day in NA, the thread can pick up a second flurry of posting activity the following morning, and presumably have a delayed or flattened peak if it's made late at night.
Favorited comments over time for 10,000 Metafilter posts, grouped by post's time of day.
posted by stebulus at 10:16 AM on December 4, 2011 [1 favorite]
This suggests that we should apply a time-weighting to the popular favorites algorithm so that late night comments or popular comments in 'stale' threads have a better chance of being seen...
posted by kaibutsu at 2:40 PM on December 4, 2011
posted by kaibutsu at 2:40 PM on December 4, 2011
I'm not sure my breakdown by time of day shows that comments on late-night posts are less likely to be favorited. My graphic reports just the raw number of favoriting events, not accounting for there being more posts at certain times of day, or for there being more comments per post at certain times of day. I haven't thought seriously about what the right normalization would be.
What is the popular favorites algorithm?
posted by stebulus at 8:19 PM on December 4, 2011
What is the popular favorites algorithm?
posted by stebulus at 8:19 PM on December 4, 2011
Back when we graphed site activity by hour, it looked very much as though commenting rates and favoriting rates varied in extremely close proportion to each other. It seemed like that the ratio between the two would probably stay pretty consistent in that case. But I don't remember actually checking to see if that is true before.
So here's a plot of the average favorites per comment based on when the comment was posted. This is divided up into 1 hour slices, and the smallest sample size for one of those is 86,492 comments in the 2:00-2:59am slice, so it's a pretty big sample size. The busier hours are looking at over half a million comments each. So I think the variability seen is likely to be a real thing.
Given that the number of comments varies by about 6:1 from the busiest hour to the quietest, a roughly 13% different in average favorites per comment isn't a really huge difference, I guess, but it's there all the same.
What is the popular favorites algorithm?
I think it's just the top 20 posts and top 20 comments based on counting favorites received in the last 7 days, dynamically generated whenever you load the page. So it's not that easy to determine retroactively which things appeared on this list and when.
You can't say anything with more than N favorites, or more than N favorites in 7 days. I think you would have to step through a series of time intervals and actually run the select top 20 queries for the 7 days preceeding each interval to get it pretty close.
posted by FishBike at 10:27 AM on December 10, 2011
So here's a plot of the average favorites per comment based on when the comment was posted. This is divided up into 1 hour slices, and the smallest sample size for one of those is 86,492 comments in the 2:00-2:59am slice, so it's a pretty big sample size. The busier hours are looking at over half a million comments each. So I think the variability seen is likely to be a real thing.
Given that the number of comments varies by about 6:1 from the busiest hour to the quietest, a roughly 13% different in average favorites per comment isn't a really huge difference, I guess, but it's there all the same.
What is the popular favorites algorithm?
I think it's just the top 20 posts and top 20 comments based on counting favorites received in the last 7 days, dynamically generated whenever you load the page. So it's not that easy to determine retroactively which things appeared on this list and when.
You can't say anything with more than N favorites, or more than N favorites in 7 days. I think you would have to step through a series of time intervals and actually run the select top 20 queries for the 7 days preceeding each interval to get it pretty close.
posted by FishBike at 10:27 AM on December 10, 2011
13% different in average favorites per comment isn't a really huge difference, I guess
It seems possible that even a small difference in distribution could make a big difference in the probability of making it onto a "top 20" list.
I thought I'd try simulating the Popular Favorites page (for posts only, not comments) using the favorites data for 2010. (Code here and here.) This is more than a little sketchy, so take the results with a grain of salt, but: Probability of a 2010 Mefi FPP appearing on a simulated Popular Favorites page, by post's time of day.
posted by stebulus at 9:40 PM on December 10, 2011
It seems possible that even a small difference in distribution could make a big difference in the probability of making it onto a "top 20" list.
I thought I'd try simulating the Popular Favorites page (for posts only, not comments) using the favorites data for 2010. (Code here and here.) This is more than a little sketchy, so take the results with a grain of salt, but: Probability of a 2010 Mefi FPP appearing on a simulated Popular Favorites page, by post's time of day.
posted by stebulus at 9:40 PM on December 10, 2011
You are not logged in, either login or create an account to post comments
posted by Plutor at 9:24 AM on December 2, 2011 [1 favorite]