In light of melative, Rating
I was lead into a post over at Meaty Anime Blog recently, initially by the basis (Ouran and Towards the Terra), but something was stated:
…it’s the twelfth top rated show on MyAnimeList. Not that the last point means anything,…
This may or may not be out of context, since I cannot attest the same reasoning for the statement, but it has implications… ratings are flawed. And no, this isn’t a MyAnimeList-only problem (see how many 10s are given to popular movies on IMDB, ie Rush Hour 3).
What we have is an arbitrary rating system with a limited-number scale. In other words, I can arbitrarily rate multiple items 10/10 as much as I want…. but does this mean they are equivalent? Definitely not. Perhaps we should increase the arbitrary scale, as AniDB does with enhancement scripts (test that thing for steroids!), from 10 to 1000, does that make it better? Probably, but it also increases the level of “pain-in-the-ass”. Do I sound tempered?
This does not irritate me, because I don’t have to use it if I choose, though I would like to see accurate ratings, and so I come to the point, melative. (I know, I’m a broken record, shoot me for being my own fanboy… jkwut or just jk?)
The general idea: Relativity
melative.com has an embedded RRS (Relative Rating System), a premise which has held with me for a few years now (apparently there is a book that has a similar notion called slicing, I don’t know that the book is titled though). The idea is extremely simple, but it can work on the large scale, via recursion. Let’s do an example:
I have 2 media, A and B, let them be on the same “tier” or “level”. In this state they are equivalent, I feel they are equally great. If I feel A is greater than B, I place it on the next level up, or if I feel B is less impressive than A, I place B a level down. Effectively, I have rated A and B relative to each other, clean, concise, simple.
Rinse and Repeat
When dealing with a large list of titles, this may seem like hell, but it always uses the base… A=B, A>B, or A<B; there are only 3 options. On the large scale, it is easier to know a fuzzy location of a title, perhaps in the top 10%, great melative knows how to do that too, but that isn’t the point. Just because the ratings model is a recursive base case, does that really yield accurate averages for a title?
I ask too many questions, for more information, here is an overview/instructions on this RRS, but the money is in the Theory section. Douzo meshiagare!
Ryan A
note: further discussion, questions, or ideas can be directed to me via e-mail or on irc (#animeblogger or #melative).
May 7th, 2008 at 12:38 pm
That sounds like a cool way to rate titles- actually it might be easier than thinking about a number score. I also like how anime news network did their stats/rating list. They have a bunch of different ratings, and they made lists of most “overrated” or “underrated” that was pretty interesting (based on comparing how many votes and the scores of the votes). http://www.animenewsnetwork.com/encyclopedia/ratings-anime.php
May 7th, 2008 at 4:50 pm
That makes sense and I might try that in a non-automated way once I start watching more things. I don’t think I even have enough to effectively compare, looking at some MAL users. That’s a lot of stuff watched….
May 7th, 2008 at 6:03 pm
Susie Q, Yea, that is one plus, scoring underated/overrated. I remember checking that out a couple years back. I am not sure these are needed with the melative ratings, but I will probably consider the way under/over rating would occur on melative (the point of the RRS is to eliminate such things) ^^
Caitlin, well, the good thing about RRS is the minimun is 2 items ^^
May 10th, 2008 at 10:22 pm
It’s a good idea, strengthening your thought that everything is relative.
May 11th, 2008 at 6:24 pm
After my recent ethics course, I can’t say everything is relative ^^, but at least by ordering media in such a way, it can be made relative… or something like that.