Hmmm, interesting.

Edited to add: Didn't understand at first, then saw the profile comment which helped alot. Thanks, Lucky!

Think of it as being whatever SiftBot thinks is a method by which popularity may be determined. To be honest, we basically just told SiftBot our thoughts and he went off into a dark corner and asked not to be disturbed.

I'm kind of curious how it works myself, but the last time I asked he gave me an icy stare that chilled me to my very core. I fear SiftBot.

I made the mistake of disturbing SiftBot in his dark corner, he threw what appeared to be some old chicken bones at me and muttered a Hatian curse under his breath. I haven't been sleeping right for a week, some things are better off left alone.

thanks MLX, put in a message telling people to check their profiles and linking to the two pages in question in case anyone else was confused.

Just to let you know, comments to this post are showing up in recent comments even to users who aren't in the top-secret whatever it is.

I think you just got yourself an invite gorillaman, welcome to the beta group.

What's the scale? 1 to 7? I wonder if a more graduated system (to take it to an extreme, 1 to 100, or perhaps something a bit smaller) would be even better because it would offer more "feedback" to users, allowing them to see a bit of the effect from their vote (doing a vot eand seeing no change in siftscore may frustrate some) or perhaps the mystery is part of the appeal. Just brainstorming here on the psychology of sifting.

Are we secret now?

BTW, if you're inviting anyone else, a direct link to this post would be nice for the lazy among us.

BTW, if anyone looks at my favourites, will they see that this is on my list?

Oh, I see it goes up to 23.. 70? Oh, does it go all the way to 100, then?

New suggestion, then: will it be possible to do a "live update" of the sift score after voting? Right now reload after voting is necessary to see the change.

three comments:

(1) i'm confused as to how this works.

i understand you're afraid of siftbot, but crazy addicts will want to understand how this works. even if it's complicated, can you explain?

(2) it's not intuitive so far

i just gave the 10th vote to get farhad's robot guitar video published. but when you go to the front page, it's not there (some osama bin laden / family guy clip is first). when i started writing this, it was on the third page. a couple minutes later, it's on the second page, even though it still has the same number of votes and the same sift score.

so there seems to be a significant random element, or else the activity from the last X minutes is heavily weighted.

the problem with that model is that the published videos will jump around so much that sifting sequentially through 16 pages of videos (current number included in the hot sift) will be confusing. if it takes 5 or 10 minutes to get from page 2 to page 3, videos may pop down the list and "reappear" on page 3, or pop up the list and be missed entirely.

which means it will impossible even for hardcore addicts to know when they've managed to cast eyes on everything published. and there will be no way of knowing when you've "caught up", ie. seen everything published since last time you checked.

(3) i don't think this is a necessary innovation, in part because it's not adding much new beyond the "top videos -- 24 hours page" to the stereotypical superficial visitor (who looks at the front page or two and the top 15 sidebar, but not much else).

right now, here's how the hot sift stacks up against the top videos -- 24 hours page:

-- osama / family guy (#1 hot sift, #2 top 24)
-- restless goalie (#2 hot sift, #6 top 24)
-- beck loser (#3 hot sift, #1 top 24)
-- simpsons epcot center (#4 hot sift, #4 top 24)
** pedophile apeshit (#5 hot sift, #17 top 24)
** simpsons wiggum twin peaks (#6 hot sift, #19 top 24)
-- strangest tourist attraction (#7 hot sift, #9 top 24)
** family guy creationists (#8 hot sift, #16 top 24)
-- radish seed timelapse (#9 hot sift, #7 top 24)
-- silence of the lambs trailer (#10 hot sift, #8 top 24)

only 3 videos are on a different page than they would be, all with >10 votes in <10 hours, and thus all certain to be top of the 24 hour page before long (as older videos disappear)

Comment hidden because you are ignoring dag. (show it anyway)

Thanks for the meticulous feedback PM - what we are trying to do is include other metrics besides votes to indicate a post's popularity.

From out perspective, as we watch our traffic meter climb - we are realizing we need more tools to help us deal with the dreaded "queue rot" and 20 videos per hour being posted to the front page.

I take your point about the volatility - but I think this might work itself out over time, if we evolve this model, in theory, the higher the traffic the less volatility because you are looking at aggregate indicators across all viewers.

sorry, got cut off...

if the goal is to give first-time or casual visitors a look at "what's best" or "what's hot", why not just make the top videos of the past 24 hours page the default front page?

otherwise, why not use this algorithm to replace the top videos page? (and/or make that the default front page).

just don't get rid of the chronological version of the front page (even if it's hidden beneath a tab somewhere).

Attention gold star techno-ravers - just wanted to point out to the audience assembled here a mostly unrelated fact:

I have turned my private "Videostopromote" playlist public with the intent of encouraging more participation and sharing it for use among a wider audience. I invite all to make suggestions to the playlist! http://www.videosift.com/playlist/Krupo/Videotopromote

I say "mostly unrelated" because I'd like to see how *old* videos with a SiftScore of "1" will do. It's a sub-experiment to see how Siftbot will rank promoted videos.

For the course of the beta test, I won't remove promoted videos from the playlist to allow anyone interested a handy way of tracking how promoted videos are doing. Please fill up the playlist with a few more suggestions


Sidenote: agree with pm, it'd be nice to get, at the very least, a vague idea on how the score is calculated.

Also, looks like the new system is temporarily offline - I assume behind the scenes tinkering is at work.

I was really curious to see how a video with 8 votes in half an hour ( http://www.videosift.com/video/Rancid-Time-Bomb ) was going to do...

it's not offline

Weird, I can't see the siftscore... (?)

Comment hidden because you are ignoring dag. (show it anyway)

We're still playing with the SiftScore bit - and didn't want to confuse people too much.

Due to unknown reasons, SiftBot has deemed the SiftScore undesireable information and has apparently removed it permanently.

To clarify the how this new system works:
Videos are submitted into the UnSifted area where good videos are mixed with crap videos. Once a video has become "popular enough" (criteria yet to be established), it is granted a permanent home on the site, initially listed on the Sifted page in the position appropriate for its popularity. The more new and popular a video, the more toward the front it will appear. Over time it will drop back from the front either from old age, decrease in popularity, or both. This process may happen over the course of hours and rarely, if ever, in a matter of minutes, as some fear.

I don't know what to make of the queue yet, but my first impression of the front page is I really like it.

I am worried about videos getting published into obscurity, immediately dropping into the middle pages and not getting any more views. How about giving just posted/published videos a temporary bonus to their score so they can have their chance on the first page of their respective area. Then maybe they'll get the votes to stay there and maybe not.

Also PM, we were planning on keeping the current front page set up as a setting under top videos. You would be able to list them the way they are currently listed on the front page.

Would it be possible for a user to set a default option? Perhaps one might want to see the queue in chronological order, perhaps another might like to see it based on popularity....if not, at least have 'chronological' available as a sorting method.

Also, I think it's important to publish the established popularity criteria. Now that we can't see the siftscore it's pretty much a crap shoot.

james roe: we were planning on keeping the current front page set up as a setting under top videos.

that's good, thanks... even if it takes an extra click to find, it'll be worth it.

dag: what we are trying to do is include other metrics besides votes to indicate a post's popularity.

does this include comments?

does this include comments?

We are not sure exactly how Sift Bot conjures this bizarre amalgamation of the popular.

MLx: "I think it's important to publish the established popularity criteria."
Theoretically, you may be correct. In practice, however, SiftBot shall never publicize how he determines what's popular because if he were to do so, everyone would try to manipulate the site to force their videos upward.

The goal of submitters should be to simply post a video that they think is quality, not try to manipulate/exploit the system to make whatever they post as popular as possible.

*Siftbot's Criteria Explained:

Viddies are randomly chosen by tag words, to establish the criteria for eval.
They are then internally discarded, then resurrected in order for siftbot to make a determination, as the whether or not the viddie has been properly discarded then re-instated.
Finally, it is sent to an internal queue, of an un-named body of sifter's, the members and numbers of said panel, calculated by number of users on-line, during the resurrection period.
Siftbot then changes hats, and acts as judge and executioner, of any posts not receiving a resurrection.

Now you all know.

I absolutely agree with your last statement, lucky, however by not publicising the criteria won't that be making suspect the choice itself? There are those who might say that the site administrators (siftbot) are working some secret guerilla marketing agenda paid for by youtube, or myspace or the Cartoon Network. How can we defend our policies if we don't know what they are?

I'm just playing devil's advocate here...sorry.

Sifters who have a larger number of listings per page will have a larger number of vids on their front page(the new one), right? Or will there be a finite number of vids on the front page?

SiftBot in a dark corner sounds a little "Blair Witch"? She isn't lurking around here, is she?

*waves* just got home from work ... trying to wrap my head around this ... but i'm going to go to the mall instead! i will think about all this while looking through clothes!! LOL

i'm definitely interested to hear what the criteria is going to be that partly accounts for a video's popularity.

what sort of time frame are we looking at before the inception of this new feature?

The secrecy is both cute and understandable, although mlx's point is valid to a degree too.

There's a finite number of variables affecting popularity - one issue given the secrecy, though, is how we can tell if "it's working", at least during the beta stage? With little more to go on other than our delicious speculation, the "light touch" theory seems apt:
http://www.videosift.com/video/Futurama-Bender-as-God

I'm going to throw out some ideas, and re-iterate a guess of what siftbot might be "into".
- votes over time - easy
- comments - some permutation, again probably including frequency over time
- age? <- I'm still curious to know how the promote function will interact - will it be an AUTOMEGABOOST to front? It should be - we really need clarification from siftbot on this point (*-promote), even if all others are ignored, otherwise that function may just as well be disabled - individual voter composition - no one's mentioned this, but I can only imagine how delicious a factor this can be: it would actually give REAL weight to people who vote intermittently (or heavily, as the case may be). This is something I would heartiliy agree should have a justificably secret algorithm, should something so wild be introduced, because it would A. throw off the whole 1 vote = 1 vote mechanism we've had to date, and B. make people feel incredibly self-conscious about their voting patterns (not voting up/down enough, or doing so too much, etc.). Additional variables could include power of vote as a function of votes casts vs. videos sifted vs. comments made

I suspect that last one might not be a full-fledged item (or may not even exist) simply because of the potentially horrendous processing power a sufficiently unnecessarily complicated formula could suck off the server.

Having said that, voter composition vs. all other factors would be an ever-present possibility.

Association with channels favourites, playlists, collectives and even mincing about with tags are other possibilities (give more/less weight to video based on popularity of tags alone).

Once the new system becomes public, it'll be in siftbot's favour for as many of these variables to be considered, because the more variables you give, the greater the futility of an exercise in 'breaking the system' through datamining or other variance analysis (what's the name of that cover in statistics? It's not ANOVA... it's the line that passes through the most points in something like a scattergraph?)... anyway, I'm running off an a second year university stats course tangent, and my mark in that course was hardly high enough to justify a long winded dissertation so I'll just reiterate my point: http://www.videosift.com/video/Futurama-Bender-as-God

So aside from discussing the algorithm, can you fine Techno Ravers (plus gorillaman) share your thoughts of how it's been working thus far?

Have you been using the test pages at all? Of particular interest are things like:
- how often do you view the test front page and queue?
- how do you feel about the videos on them, w/regard to quality, rate of up/down movement, and staying power?
- what is it like for you browsing through the pages?
- how do you find the test queue date range for browsing? do you modify it much?
- etc...

Please try to make browsing those pages part of your normal sifting routine. (FYI, we were going to just replace the old front page and queue with the test ones to get your full sifting experience out of them, but opted to just add them in parallel. Please don't give them the old once over and be done with them. We really need as much input as possible so we can determine if and how well it's working.)

Thanks, friends.

Comment hidden because you are ignoring dag. (show it anyway)

When I'm at my day job - unfortunately I don't have time to sift the 100 or so front pagers a day. (yes I'm kvetching about our queue escape level)

I really, really like to be able to see what the active posts are for the day - and this system identifies them effectively.

i understand why you want the method to be secret, but it's kind of frustrating at the same time. for one thing, it's hard to help if we don't know what's happening behind the scenes.

"popular" is the wrong word for it, "hot" or "active" is closer. you seem to be giving points for triggering downvotes and gaining comments. so far, there have been a lot of controversial videos (9-11 conspiracy, osama on family guy, the masturbating dog, even the avant-garde metronome music) higher up in the hot-rankings than simple "upvotes-per-hour" math would suggest.

i hope the new focus on rewarding "other metrics" (besides upvotes) doesn't end up encouraging flamewars, comment spam, and shock-value videos. the site has become much friendlier to Obscure and Artsy content over the last few months, and i hope that doesn't get squashed under a pile of masturbating dogs.

probably lots of people here are cheering for more dogs.

anyway, your target seems to be casual viewers, and the goal is to consistently give them more high-impact content up front. for people who click past basically every video posted to this site, it's more efficient to go chronologically, since that way you don't miss anything and you never see the same thing twice.

as a pretend casual user, i think the /hot pages are turning over at a decent rate, but not too fast. they're easy to use, and i probably don't care why videos are in any particular order. finding the front page full of "hot" content is going to hold my attention better than a bunch of random weird things that i probably don't care about.

as a serious VS user, the main benefit would be having a /hot queue where older videos with 8 or 9 votes don't get lost. right now, the 72-hour queue has nothing older than 24 hours in its first 6 pages (lamb video for "gorecki" has 9 votes in 24 hours and is on page 7), so that's not quite happening yet. suggest you'll have to re-weight the queue regarding recent activity to adjust for that.

meanwhile, the family guy "dumpster baby" song is still on the front page of the hot queue after 9 hours with only 4 votes, just because it's offensive enough to have earned two downvotes (so it's 6 up, 2 down). maybe if it gets another downvote or two it'll hit the top?

...no!

downvoted it and it dropped 3 pages in the hot queue.

sorry lucky, it's actually a better family guy clip than most of what gets published here. the only family guy stuff i really like is the broadway style singing and dancing...

gotta love musicals.....

I've been playing with the drop down boxes in the new queue and wondering if the time frames are accurate...

Memphis Soul Stew currently has nine votes, posted 1 day and 15 hours ago but doesn't show up at all in the 48 hour range. I did find it at #103 in the 60 hour range.

I also notice that the top 10-15 'HotQ' videos remain the same in each of the ranges, no matter how many votes or comments they have recieved.

I also noticed the old queue shows 213 videos, where the 96 hour HotQ shows 197.

Things I like so far about the new system:

1) Easy access to the most recent queued/hot videos

2) Doesn't seem to impede those videos that will inevitably rocket to the top (though I worry that the randomization will flatten out top performers into 40-50 vote videos rather than 60-80 vote videos)

Things that may bother me:

0) Don't like the design of Sifted/Unsifted button right now. May be an aesthetic/difficulty adapting to a new school thing.

1) Word-of-Mouth videos that struggle in the queue (but eventually make it) are going to be swallowed by the new system, UNLESS there is a metric keyed to placement in playlists, saves, comment whinging from rickegee et al about the struggle of the little esoteric video.

2) What will happen to promotes? Will they 'stick' on the front page for a while?

3) Will this affect my inexorable drive to the #1 position?

4) I fear a front page of only Family Guy. That stuff is such excrement.

more later. i need to experiment . . .

So far I'd love the system to be the default for the frontpage and the queue system. Since they are representative of community wide activity on sifts.

However I don't like that stuff that that is clearly of zero interest to most sifters is held up high due to downvote/upvote activty. Some is relevant but others isn't, there is too much emphasis on a few downvotes for the system to push up sifts.

Lene Alexander am looking at your boobs here.

So am for it replacing the current thing but would still want a user ability to change the layout via registration, this could act as a membership incentive.

I think that php meddling unsecreted this post.

thanks djsunkid, welcome to the beta, please provide some feedback on the pages themselves.

Will we still have sort options in the queue? least votes, most votes, oldest, newest

Will we still be able to see only "P" probation sifters?

Will we still be able to avoid seeing vids we have already voted on?

*discard

Discarding this post (discard called by gold star member James Roe)