Difference between revisions of "YTMND Front Page Algorithms"
m (→Suggested NEW Front Page Algorithms)
|Line 53:||Line 53:|
'''Users who liked this site also liked..'''
'''Users who liked this site also liked..'''
This isn't something for the frontpage, but rather the site profile/comments page. What it would do is take everyone who voted a 4 or 5 on that site and look at their vote histories. Whatever sites have the highest average rating
This isn't something for the frontpage, but rather the site profile/comments page. What it would do is take everyone who voted a 4 or 5 on that site and look at their vote histories. Whatever sites have the highest average rating that group(with a minimum of 10% of the group having voted on it) would be shuffled and shown in this box. [[User:2nd Piston Honda|2nd Piston Honda]] 20:35, April 24, 2007 (CDT)
== Suggested Changes to Front Page Algorithms ==
== Suggested Changes to Front Page Algorithms ==
Revision as of 01:28, April 25, 2007
This is a work in progress containing algorithms for the content boxes on the front page, while this information could be used to exploit the system, it's not too hard to figure out on your own.
Current Front Page Algorithms
"Worthwhile YTMNDS": YTMNDs with a score above 4.0 and a number of votes fitting between 4 and 425.
"Up and Coming YTMNDS": YTMNDs created within the last 48 hours that have a score above 3.0 and a number of votes fitting between 3 and 300.
Some users have expressed disdain at the fact that once a site reaches 300 votes its removed from the Up and Coming list.
"Top Viewed YTMNDs": YTMNDs sorted by views of the time segment (today, this week, all time, etc) with a score filter to filter out horrible sites.
Suggested NEW Front Page Algorithms
Write new algorithms here.
"Up and Coming YTMNDS":
Vote Limit = (300 * (Score - 2.5)) This way, a site rated a 3 (meh) would have to get 150 votes to drop off, 3.5 (pretty good) would need 300, 4 (nice) needs 450, and a 4.5 (amazing) would need 600. Just dropping thoughts, and it can be tweaked to get the vote limits you want. Stevetheninja 20:44, April 15, 2007 (EST)
Vote Limit = (20 * Score - 5) * (192 / Hours Since Creation) This is my modification of Stevetheninja's algorithm. It addresses the amount of time the site has existed. The vote limit decrease with time. A 4-star site after 1 hour would need 14,400 votes to get removed. After 3 hours, it would need 4,800 votes. After 6 hours, it would need 2,400. After 12 it would need 1200, and after 24 it would need 600. By the time it reaches 48 hours, it will take 300 votes to get it off the Up & Coming, which it will most likely have gotten. This replaces the current sharp cutoff of 300 votes or 48 hours to a more logical variable limit. EricJ2190 23:36, April 17, 2007 (EST)
Max, have a section of your favorites sites of the week or something, Max, have a #wop section, Max, put your The truly random YTMND Creator, on the fp so I don't have to open up your stupid old post. --Black-Llama 19:22, April 15, 2007 (CDT)
(Moved to YTMND:Weighted_Voting actually, related to both) This is basically an independent 'axis' of voting, by giving out 'tokens' granted to users in limited numbers..
Perhaps some algorithm combining into one box the sites' votes, # of views, and # of faves over some period of time. As far as i know, these 3 principal 'scores' of YTMND are now only conted separately.. Skullers 02:29, April 17, 2007 (CDT)
You could look at second order information (in newtonian physics, distance is first order, then velocity (distance/time) is second order, and acceleration (velocity/time) is third order). In this case, distance could be replaced with views, votes, favorites, or some combination of the three (similar to Skullers' suggestion above). This is something that can be rudimentarily accomplished right now by looking at a YTMND's change in standing from its position today vs. yesterday vs. a week ago, etc. But a more discrete and regular time period could be implemented relatively easily and the changes in rating/votes/favoriting could be traced per time frame to give a new "up and coming" vs "dropping off the face of the earth" vs "holding steady" based on how you want to threshold the velocity. Then you could take it to the third order and look for acceleration changes (changes in the velocity over time). Finally, each YTMND detail page could carry a small graphed version of these values so individual YTMND stats are also improved. KazRiprock 02:21, April 19, 2007 (CDT)
Sometimes a site comes along and doesn't follow the normal conventions of viewing/voting/rating/comments/etc. With ALL of the YTMNDs created, there is a distribution of how many comments/votes/views the "average" YTMND generates. By dividing the total number of votes (or whatever value you want to test -- call it the sample statistic) by the total number of YTMNDs, you can generate the average (mean value) and the standard deviation. Depending on the shape of this distribution of the sample statistic, there are different ways to determine the "rare" YTMNDs (ones that don't have a sample statistic close to the average). For example, if the rating per YTMND has a normal distribution (aka bell-shaped curve) then anything greater than 3 standard deviations from the mean will be rare (0.27% of all YTMNDs will have ratings either greater than or lesser than 3 standard deviations from the mean rating). These are Stand Outs that could be highlighted (the best and worst rated, the most or least votes, most or least comments, etc). While some of these categories are somewhat available by sorting a column and some of the YTMNDs are obvious because of how far from the "average" YTMND it seems that they lie (like the views for the Telemarketer YTMND), this is a way to root these designations in statistics and provide a much easier way to classify these stand out YTMNDs. KazRiprock 02:21, April 19, 2007 (CDT)
This isn't quite an algorithm, but as long as we're on the subject of front page boxes I thought I'd suggest it. My Fives would be a box that shows a random sample of the sites that you, personally, have given five stars to. Basically, it's a quick, dirty portal to what YOU consider the best that YTMND has to offer. This does offer some overlap with user favorites, but it has a few benefits that faving a site doesn't: (1) encourages people to VOTE, and vote honestly, and (2) brings back from the depths long-forgotten YTMND's that we all voted for before favorites were implemented. The concept could be expanded further with a multitude of "My Ratings" boxes like My Fours (for the good but not great), My Ones (the bad) and My Threes+ (threes, fours, and fives all in the same box), etc. DXSSI510 20:00, April 20, 2007 (CDT)
Users who liked this site also liked..
This isn't something for the frontpage, but rather the site profile/comments page. What it would do is take everyone who voted a 4 or 5 on that site and look at their vote histories. Whatever sites have the highest average rating from that group's votes (with a minimum of 10% of the group having voted on it) would be shuffled and shown in this box. 2nd Piston Honda 20:35, April 24, 2007 (CDT)
Suggested Changes to Front Page Algorithms
Worthwhile section would be sites that have fallen off of the up and coming in the past, not sites that only got around 100 votes. I'd suggest 4.1-4.6 rating and over 300 votes (with no cap). BTape 17:30, April 15, 2007 (CDT)
"Up and Coming YTMNDS":
Up and Coming would have 20 sites. Anything rated over 3 could be shown in the list just as it works currently. After 500 votes the site would fall off. This would essentially allow a site to be on up and coming for an entire day if deserved and would catapult the sites that fall off up and coming with high ratings right to the top voted. Top viewed does not often gain a site enough votes after it leaves the up and coming, so this way the best sites would be more likely to make top voted. BTape 17:30, April 15, 2007 (CDT)
Another good idea presented by frenchbreadpizza is to have a time limit rather than a vote amount, though I still like the idea of good sites transferring from up and coming to top voted. BTape 18:57, April 15, 2007 (CDT)
Instead of having it limited to 300 votes, have it limited to a set number of hours. If a site is on the U&C for 12 hours, it will be able to shine without sticking around too long. Let's say there's a site in the U&C that truly rocks. People will flock to it, upvote it, love it, and it will hit 300 votes in no time. Now let's say there's some lame Nintendo site or some other fanboy thing. Most people won't even bother to view it, meaning that it will take much longer for it to achieve 300 votes. And it, in turn, will spend more time in the Up And Coming than the site that actually deserves it. Giving them a set time limit will put them on equal ground. In that 12 hours, the good site could get thousands of votes, and the lame fanboy site could get 100. But they both had the same time upfront, they both had the same chance, and they both got the most they could. Everyone wins. frenchbreadpizza
First, this is all my opinion, which may be completely wrong, and should not (please) be taken as anything near fact. But as BTape says, raising the Up and Coming vote dropoff would probably be a welcome addition, for more than one reason. First, the sites could make Top Viewed if they were deserving. One may argue that increasing the limit would basically increase the number of votes on all "pretty good" sites to 500 (or some other number), and wouldn't benefit the Most Voted, but my opinion is that the extra places between 300 and 500 would "filter out" most of the "pretty good" sites (they'd fall off after 300 but before 500 to better sites) to leave the exceptional sites to take the spotlight. Stevetheninja 19:04, April 15, 2007 (EST)
You could just make it so that it's dependent on number of votes in an amount of time, and the value of those votes; i.e. a site that sits on the up and coming with a couple high votes but then get no votes would get dropped the same as a site that gets downvoted.
For Up and Coming, you could experiment with weighted voting. Users whose votes tend to be closer to the average for a YTMND would have their votes weighted higher than users who tend to be off-base (e.g., upvoters and downvoters). That way you get the most leverage out of the first few votes on a YTMND and reduce the influence of gimmick voters. Or, if you want to get even more leverage, you can take into account information about how much a user tends to vary from the YTMND average, and by how much. So if a user tends to rate YTMNDs 1 star lower than the average, you'd add one to their vote when computing the average. One good way to do this would be a simple linear regression: [YTMND Average] = a0 + a1 * [User's vote]. Once you've estimated a0 and a1, you just plug in the user's vote on future YTMNDs to get an estimate of the YTMND average. Perhaps you could also use the R^2 value from the regression as a weight (conveniently, R^2 is always between 0 and 1), which would take into account information about the variance of the user's votes from the YTMND average -- so erratic voters would have less weight than reliable ones. Re-weighting users each night would save processing. I'd be happy to think more on this or talk about how to implement it in a relational database; I'm getting my M.S. in statistics right now, so it's something I enjoy. --Sk19842 21:26, April 16, 2007 (CDT)
Mod favorites could be expanded to include all the favorites of the mods and it would then have enough content to be worth showing on the front page. BTape 17:30, April 15, 2007 (CDT)
Users must pass a standard intelligence test and be verified "Not Completely Worthless" by mods in order to have their sites get on up and coming. You know who you are. Actually, you probably don't. But everyone else does.
The nsfw panel should have sites marked nsfw, and if that site was on recently created it should be taken off recently created and added to Nsfw panel.
SonOfPrince 16:14, April 15, 2007 (PST)
"The Jenius Ideas":
Time Codes: a site that consistantly gets a high number of votes over a given period <x> is likely to be a higher quality site. If for example over a period of 30 minutes a website created in the last 3 days received 15 new votes 4+ then it could be considered more likely to be a good site.
voting ratio: the # of logged in users that view the site vs the number of them that vote >= 4 and you have another good indicator. Again this could be considered over a period of time for UaC. General idea is for good sites pepple will be more prone to voting.
10 stars: the introduction of a 10th star rating, somewhat like favourites but a person would have an expendible quantity of them, something like one 10 star vote for every 2 weeks they have been an active member (posting comments, ytmnds, voting).
Asset Thief: It strikes me that origional content that is re-used will generally be better content. A site being the first to use content that then has its assets used in many other sites could receieve a ratings bonus.
--Jenius 20:30, April 15, 2007 (CDT)
Best in topic w/ Hubs and Authorities:
- I don't have much time to explain, so search for "Hubs and Authorities"
or look in the Cornell INFO 204 course notes(it's not available publicly - the cite is Peter Coy, Business Week, 6 March 2006) for details. The authors were thinking of using websites as both hubs and authorities, but we have it easier.
- Authorities = lists of favorites, or list of 5-star ratings for some user, etc.
- Hubs = YTMNDs
- Get a keyword. i.e. Let a user input a keyword, or have a predefined/precomputed list of popular keywords or fads. The second is probably more feasible.
- find a bunch of YTMNDs (the naive way) and lots of favorites lists containing some of those YTMNDs
- YTMNDs with lots of incoming pointers = higher hub score (ex. score = # of incoming pointers)
- Authorities (favorites lists) with lots of high-scoring YTMNDs = higher authority score (ex. score = sum of hub scores pointed to)
- Recompute hub scores (YTMNDs with lots of high-scoring authorities pointing to them = even higher hub score, ex. sum of authority scores of those that point to it)
- divide by something to keep the numbers sane
- repeat 4-6
- Take the top x and display them
- Could be improved on by using some sort of clustering mechanism instead of a keyword, and then labeling the cluster afterwards. See http://expertvoices.nsdl.org/cornell-info204/2007/03/01/networks-that-cluster/ where I badly tried to explain one that I found in Nature mag. The only problem - how to define "similarity"? Something involving favorites lists again?
AySz88 21:05, April 15, 2007 (CDT)
Users that have an average site rating of 4.0 (or whatever you determine) with a minimum of 20 sites (again can be changed) are featured in this main page tab. Sites that have been created in the past 3 days by these users will cycle thru this tab, randomly showing any 5 eligible sites on each refresh. How often do the "Ytmnd superstars" produce a crappy site?
- This section could also require other users to "vote in" Featured Artists. Each account could have X ammount of "Artist Favorites" that could be used similiar to the Favorite site vote--if a user receives Y ammount of Favorite votes, there sites are eligible for the Featured section
--Ryuudo 20:40, April 15, 2007 (CDT)
Registered Users Only
I don't like the idea of making ytmnd only for registered users. Half the fun of the site is finding a ytmnd that everyone can enjoy, and pasting the URL into an msn conversation. For example, anyone familiar with... television? can enjoyAlso, it's registered users that contribute to the fact that good content is constantly being overlooked. It's too easy to make a good site that gets a few votes, but is knocked out of the "recently created" list before it gets the recognition that it deserves (happens to me all the time! :^)). Making both the "recently created" and the "up and coming" boxes greater than ten sites (remember the good old days of the top fifteen?) could help combat this.
--- SevenAteNine 22:05, April 15, 2007 (EST)
Only registered users can vote and give your recently created site the "recognition that it deserves," so the only hit you'd take if only registered users could see your site is the number of views it achieves, which would also be felt by every other ytmnd. It'd be a level playing field. Though it's not the only solution, Max, Registered Users Only is a step in the right direction when it comes to cutting bandwidth costs.
--- ghosttown80 7:30, April 16, 2007 (EST)
Trial By Fire
Sites with less than 1,000 views but over a 2.5 or 2.0 in rating. Maybe sites that fail the trial by fire and get lower ratings than they started with get deleted. --Flakmaster 00:25, April 16, 2007 (CDT)
Make a Front page algorithm with Classic ytmnds with a minimum rating of 3.5.
Up and Coming YTMNDs
To show up in up and coming, a YTMND should have the following:
- rating > 3
- NOT IN TOP 5 VIEWED TODAY
- Number of views must be increasing (Views Yesterday - Views 2 days ago > 0)
- Score must not be decreasing (Rating yesterday - Rating 2 days ago > -.2) [the -.2 is to allow for a slight decrease)
This way, ones that people get sick of (the views decrease because everyone's seen it) disappear. Rlbond86 15:12, April 16, 2007 (CDT)
My Option 1, A fix using the existing types of boxes
To me, ever since around the creation of the U&C, YTMND has become way too fast paced. There needs to be some sites on the frontpage that stay there for a while, allowing fads to be born and a culture to form. Like in the old days, good sites would stay on the top 15 for several days to a week. I know we have the Most Voted On, but i don't think the content is good enough there and it's not enough.
Here's the version i would probably implement today if i had control of this site.
Default Frontpage Layout
- Recently Created
- Up and Coming - Top rated sites created in the last 48 hours with a vote total of between 6-299
- Top Rated This Week - Top rated sites created in the last week with a vote total of 300+
- Top Viewed This Week - Top viewed sites created in the last week
- Worthwhile - Sites with 4+ rating and at least 50 votes
- Mod Favorites
- Random YTMNDs
- Sponsorships This Week
Notice, i say "created in the last..", which is different than it works right now. Right now if you sort by Top rated This Week you'll get sites that were created long before the last week, because it's only sorting it by 'votes in the last week'. In the system i'm proposing, it's excluded from the lists if it's over a week old. Also, no other parameters should be used other than the ones listed above.
Here's how i would set it up
- Left side
- (5) Sponsorships This Week
- (10) Up and Coming
- (10) Top Rated This Week
- (5) Random YTMNDs
- Right side
- (10) Recently Created
- (5) Worthwhile
- (10) Top Viewed This Week
- (5) Mod Favorites
The way this works is that a site is submitted, then if it's good enough, it goes to U&C, then after 300 votes it can go to Top Rated This Week, then if it stays there long enough, it can get on the Top Viewed This Week. After a week, a site can look to be seen on the frontpage in the Worthwhile or Random sections. The Random i added because i think it's important for users to know that no matter what, all sites have another chance at success and none are absolutely shut off in the abyss.
The Sponsorship section i reduced to 5, because yeah it's a good idea, but it is just an idea that was an added part of the site..it shouldn't dominate the frontpage as much as it does right now.
That's about it for now. This is just a recommendation for what you can do right now, without any entirely new types of content boxes or before you add them. 2nd Piston Honda 15:27, April 16, 2007 (CDT)
- Now that i'm looking at it more, there are some problems with this version, like the need to have a daily pulse of what's being viewed alot on ytmnd which the Top Viewed currently serves, and what this different understanding of "top rated/viewed this week" would mean for the rest of the /browse lists. The role that my "top rated this week" fills is definitely needed though, a ratings-based list with 300 min votes that sites can go to after leaving the U&C, so you might just need to give it a different name. My "Top viewed this week" should probably be replaced with the Top Viewed as it currently is. 2nd Piston Honda 16:00, April 16, 2007 (CDT)
Eliminate All Shit Sites
Any site under 2 stars automatically gets deleted after a specified period of time... Since the renewal time limit is 72 hours, it would be reasonable to give it 4 or 5 days after its creation. It would clear out all the bullshit people post but never delete despite how bad it is.
This system would allow for huge relief for the servers' memory without ruining any meaningful content of the site.
Customizable Front Pages
Since the new site layout, adding and removing content boxes has become extremely easy, so I was thinking on a method of customizing a user's Front Page. There is only a limited amount of space on the Front Page, yet there are tons of different kinds of Content boxes that could be put up. So why not give registered users the ability to choose which of those Content boxes they want on their Front Page. For Example, I could choose to have Worthwhile YTMNDs, Up and Coming, Top Viewed Today, Top Viewed This Week, and Top Viewed Ever...but someone else could choose that they want to have Recently Created instead of Top Viewed This Week, or Spondored YTMNDs, etc. Of course, there would be a Default layout for unregistered users. Foolmoron 15:40, April 21, 2007 (EDT)
A Leveling system could be used to determine how much a user's vote counts, and could be used with the Weighted Vote Algorithm that is being worked on. It could be a system similar to Newgrounds.com's leveling system. Something like every comment posted gives you 1 Experience point, but if that comment is voted down to a negative comment, it would subtract 1 Experience point. Also, every YTMND submitted with a score over 2.0 would give 2 points, 4 points for over 3.0, etc. But if the YTMND is below 2.0, you lose 2 points. With enough Experience points, you would level up, and increase your voting power by a certain amount. These are all suggestions and could be changed at any time, but I think this sort of method would work well with the YTMND community. Foolmoron 15:47, April 21, 2007 (EDT)