Donmai

Score weighting?

Posted under General

I've heard some complaints that explicit posts get an unfair amount of representation in the popular pages. Some statistics: the average score for explicit posts is 14.8 (stddev 16.33), for questionable posts it's 9.80 (stddev 11.18) and for safe posts it's 6.06 (stddev 6.85).

My proposed change would make it so that voting up on an explicit post would only raise it 0.66 points, while for safe posts it would be 1.62 points. Questionable posts would remain at 1 point. Thoughts?

Updated by kumarei

I wouldn't be against it. I know I've voted down explicit posts that seemed to be in the popular pages solely due to the size of the breasts and regardless of the quality.

Is this meant to be retroactive? I ask because the easiest way I see to do this is to multiply by a factor after the fact while storing the number of votes the same as we do now, as integers. That method would also allow for a mis-rated post to be affected appropriately once it's fixed.

It's something that crossed my mind too, since I browse by the "popular" feature myself quite often.

BUT you forget that a lot (or let's say most) of the people are here to watch explicit pics.
Changing the votingcount wouldn't be very intuitive ("I voted, but it's still the same count?") and won't reflect the real popularity (since, again, most people visit the site for explicit pics).

I for one am against the voting-change. There can be better solutions, e.g. put three popular pages with additional filters (safe, questionable, explicit), so everyone can decide for themselves.
Unfortunately (if I see this correctly), the query uses more than two arguments (posted:today, order:score AND explicit/questionable/safe). Actually I think this was the same issue with the popular by month/week pages.

If you really want to treat the pics (statistically) equal the just add a multiplication factor in the popular mechanism (like already proposed). It serves the purpose of the complaints but doesn't butcher the voting system itself.

Updated

It would make sense to use this different weight for the "Popular (by *)" lists only. For individual posts, score doesn't mean much if anything. (At most, mods might make use of a poster's average score, but they'd know about this bias I guess.)

Another approach might be to have three different sets of popular lists, one for each rating. Actually, I think that'd be a good idea regardless of whether score weighting is implemented.

Frevel and Piespy's suggestions (seperate popular listings) is very similar to the way Pixiv does things, where they have one list for safe pics, and one for R-18 (explicit) ones. That also sounds like a reasonable approach.

While I don't pay much attention to score in general, I agree that porn always seems to get the highest scores, even when it's awful.

A "Popular by [time period]" of safe or questionable images only would be great. Most of the time it's clogged with porn. Usually the better porn, but still porn.

I probably approve 20 times as many safe images as explicit. And looking at my favorites, I strongly prefer safe and tame questionable images. My favorites seem to come out like this:

3,829 safe favorites
1,218 questionable favorites
87 explicit favorites, exactly zero of which involve penetration and most are borderline questionable.

One problem with weighting votes is that you'd tempt people to rate their images lower than they should in order to raise the score higher.

Anyway, the main time that votes matter for me are popular and order:score. If something could be done with those it'd be worthwhile.

Updated

I'd definitely appreciate weighting. If scores are to reflect some kind of relative ordering of posts, fixing the obvious bias makes a lot of sense. Not to mention that it irks me greatly when even the most awful porn gets higher scores than most of the awesome non-porn images. Purely psychological as it might be, it's still bothersome.

For the record, my favourites structure is very similar to jxh2154's, and I loathe almost all of the porn here. One problem with porn's prominence is that it tends to be very self-reinforcing. I've seen great many people come here with the assumption that it's a porn browsing site, not an image archive. This skews the image, introduces even more porn, and causes general moderation headaches further down the road. I think there should be more pronounced separation of porn and everything else, and I'd like to ghetto porn as much as possible. One idea I've been having is to split explicit into two ratings: porn and explicit (ie. images of greater merit than just jamming it in, that happen to be clearly NSFW). That, however, has issues with being rather subjective and very hard to define, so I don't really know how it could work. But I wish it did.

I'd be lying if I said I disliked all the NSFW images here, but I tend to find "questionable" images far sexier than most of the stuff under "explicit". Your distinction between porn and explicit would be nice, but I do agree it's too subjective to work. I'm sure people get confused about the ratings just with the three levels we have.

Really the more I think about it the more I like albert's initial proposal (and shinjidude's comments). I know I worried that people might rate images in a safer category to get a higher score but... well we can just ban them. And I do hope people should realize, the scores of your posts tend to have little to nothing to do with getting an invite. The user list tracks your percentage of positive and negative scored images, not your average score (a smart change). So trying to inflate your posts' scores won't get you anywhere.

But, I shrug. I don't feel super strongly about it either way.

albert said:
My proposed change would make it so that voting up on an explicit post would only raise it 0.66 points, while for safe posts it would be 1.62 points. Questionable posts would remain at 1 point. Thoughts?

Would downvoting posts also use the same system? And mod downvotes with their extra weight as well?

Post scores with two decimal places would be interesting, to say the least, or would they be rounded?

This wouldn't affect the scores of any images already uploaded, correct?

Anything that makes it harder for low-res porn dupes to get a higher score than the original is A-ok with me.

Also, I assume favoriting the image will affect the scores in the same manner.

Updated

I think that voting should still add/subtract 1 point for posts of all ratings. From the perspective of someone who works with databases every day, if you're going to weight them it would be most sensible to only apply this for display purposes (be it on a statistics page or the page of the post itself).

Soljashy, Frevel, et al. are right. IMO, weighting scores differently based on rating is overly complicated (and also adds a slight incentive to misrate posts). Why not just have separate "most popular" pages, or create an option to filter them?

I don't use score to compare posts' quality, I see it more as a way of finding out how many (net) people liked it enough to upvote it. This would disappear if you started weighting the votes.

And really, creating rating filters for the "popular" listing would pretty much solve the problem you mentioned, wouldn't it?

0xCCBA696 said:
And really, creating rating filters for the "popular" listing would pretty much solve the problem you mentioned, wouldn't it?

Actually, this *is* a problem. Like I mentioned in my post, doing a "popular:safe" page would mean a db query with three arguments/tags (order:score, date:today, rating:safe), but "normal" users can only search up to two tags at once.

(and don't forget the load on the server with the three popular pages with a additional tag)

A solution would be to just allow three tags with a "normal account" (and therefore three popular pages) and for the rest (who aren't registered..is it even possible to visit this site without reg?) only two tags. Those are mostly interested in the current version of the "popular"-page as it is right now anyway.

Some aggressive caching could make that point moot. Making the popular page update every ten minutes or so.

It's a nice solution for popularity pages, where instant updates aren't usually a concern.

I'm in favor of keeping the scores as they are and creating "popular by rating" pages.

Frevel said:
Those are mostly interested in the current version of the "popular"-page as it is right now anyway.

I think that's a hasty assumption. When I was a plain member, I posted and favorited >95% non-explicit pictures, and would have liked having a popular worksafe posts page.

Updated

I should hope that the "limited number of tags for normal accounts" is implemented at the search page level and not at the database connection level. If so, it'd be trivial to make an exception for the "popular" page.

Anyway, I don't know how queries are optimized for the search page, but I would assume that the popular page's data retrieval mechanism is implemented differently, to take advantage of the fact that the actual database query is static (except for the date range) - static selections as well as static quantity and sorted range of posts retrieved.

I don't know if this has already been suggested, but in response to the first post, I wouldn't change how the votes are saved; instead, I would adjust the displayed score later, when displaying and sorting, by dividing the sum of the votes on a post by the average rating in its category. Advantages:
- retroactive
- doesn't lose data
- works correctly if a post's rating is changed
- the averages can be recalculated every once in a while, readjusting the displayed score if needed
Disadvantages:
- extra calculations; the adjusted score can be cached in the db, though.

I agree with LaC and would furthermore like to make this an option/switch in the Settings. Or maybe you can do something like show both the "true" (unweighed) score and then the weighed score in a smaller font next to it, and you can click on either one and that becomes your preferred sorting method when searching by popular/hovering. I don't know if this would be much extra system usage or would be annoying to code, but I'd really like that.

EDIT: Upon further reading it seems my post was pretty redundant as others have suggested this exact thing already. Oh well!

I agree with Frevel et al. that how votes are saved should not be changed. Also, if we are going to have weighted scores, people should still be able to see and use the raw score.

Display both.
Score (raw): 66
Score (weighted): 44

Search for either.
pokemon order:score_raw
pokemon order:score_weighted

And if you did order:score, maybe it should default to score_raw, since that is the current meaning.

Update: Ah, I posted right after スラッシュ.
Update2: Wow. Apparently I did my first "Update" right after スラッシュ did his "EDIT:".

Displaying both would defeat the point of having a weighted score in the first place. Either they're adjusted or not, showing both is pointless. And it's not like the "true" score gives you any more info, you still don't know how many people liked or disliked it, how many votes were dupes, how many were mods, etc.

葉月 said:
Displaying both would defeat the point of having a weighted score in the first place.

Some people seem to want weighed scores for whatever reason; I (and some others, apparently) don't want them, also for whatever reason. It's not only not defeating the point, but actually superior if both are available for whoever prefers either one.

--My "whatever reason" for not wanting weighed votes, feel free to skip if it's boring--
I mean I personally think weighing is a really dumb idea. If I upload a picture as Safe, and someone adjust the score as Questionable, that image doesn't suddenly become less popular, but that's exactly what would happen if weighed scores are put in, giving me bad information if I casually browse my recent uploads for scores (which I do from time to time to see if certain artists/images are appreciated here). Worse still if I upload an image that's Safe as Questionable and someone fixes it, since I can't tell if the sudden jump in popularity is due to people liking the image or my mistake in tagging (especially if I don't remember tagging it wrongly--this happens a lot).

It's even worse if the vote scores are hard-coded into the system: then, I could upload an explicit image as safe by accident, people would vote it into high heavens giving it a HUGELY (damn italics!) inflated score. I'm pretty sure this is a moot point however, as I doubt that's gonna happen.

I also don't like the idea of weird points when I press "vote up" or "vote down". Being from the Internet, I'd like my +1 to be worth +1, not +0.66, +1 or +1.62, weather permitting. Even worse when it's a vote+favorite, suddenly that's worth +3 because the image happens to be safe? Or barely even +1 if the image is explicit! That doesn't make any sense whatsoever to me, and it will be even worse for the vast majority of people here who aren't concerned with the forum or danbooru-development.

But anyway, who am I to tell you guys what I think is dumb if you happen to like it? I'm almost certainly never gonna use it, but there are probably loads of features on this site I never use that people like, and that's great because there's lots of things some people want that others don't.
--End reason--

The only concession I can make on this is that displaying both at the same time might be a bit confusing interface-wise, so it'd probably be better to have it as a flag in Settings. I actually wrote that at first but then I thought it might be better to call attention to the change for casual users who don't read the forum.

Updated

1 2 3