[REQUEST] Data for testing the accuracy of a new recommendation algorithm requested
[REQUEST] Data for testing the accuracy of a new recommendation algorithm requested
Hello Lemmings,
I've been working on a new anime recommendation algorithm the past 4 months and now I pretty much have the algorithm done.
To test the accuracy of the algorithm, I am requesting any MyAnimeList user with 50+ completed animes to post their MyAnimeList username along with their overall top 3 genres in descending order. This will then be compared to what the algorithm guessed.
Thanks in advance.
Edit: Thanks a lot, guys. I finally released the project! You can find it here: https://github.com/Asudox/anote
Will algorithm be open source?
https://myanimelist.net/profile/Ludrol
Adventure
Mystery
Slice of lifeYes, it will be as soon as I am done with the website and training.Thanks for the data.
The guess is:
- Fantasy
- Adventure
- Action
So yeah, this is mostly because you watched more animes with those genres (or scored them better) than animes with the genres you specified. Thanks for participating.
I just released the project, you can find it here: https://github.com/Asudox/anote
I would be more than happy to help, but I don't have any kind of tracking system for what I have watched. It's something I have lamented in the weekly general threads before, but I just never feel up to spending the time to go back through the years and track it all.
Good luck!
Thanks.
https://myanimelist.net/profile/isyasad
Drama, Romance, Mystery
Is this anything like anime.plus? (Which, for the record, decided that my top three genres were mystery, drama... vampires)The thing is anime.plus is more of a statistic tool than a recommendation algorithm. I can't comment on their recommendation algorithm since it's written in PHP and I don't know PHP at all (my algorithm is written in Rust btw). What I noticed is that they mixed in themes to genres. Probably because of MAL's poor labeling.
I've also been trying to add some rather stalker-like factors into my algorithm just to test, such as your average watch time of 12 episode animes and changing the weights according to that. However, it became complex pretty soon and my algorithm slowed down to 1 millisecond from 500 microseconds, so yeah. It also didn't have that much of an impact and I figured out that it is rather dumb since liking a certain genre does not mean everyone will binge-watch the anime like I do.
The thing is, anime.plus is more like a statistic website rather than a recommendation algorithm. I don't really know how their recommendation algorithm works, so I can't comment on that.
your combo id: 104080
...which translates to:
- Action
- Comedy
- Drama
https://myanimelist.net/profile/Amberlie
- Psychological / Horror
- Mystery
- Action
My MAL isn't complete sadly, as I only recently started adding the animes I watched, and I added mostly the ones I liked and remembered enough I watched in the past.
That's okay. I'd also like to test it against small animelists.
The guess is:
- Drama
- Action
- Comedy
Seems like it's off a lot. Though this is partly probably because you either scored the animes with those genres more than the ones you mentioned or you just recently started liking these genres you listed out. Objectively looking at your animelist points out that it is mainly of animes with drama, action and comedy.
Are we talkimg mal genres which are lacking or any? I haven't watched anime in a while but I've been a member since 07 and have watched hundreds of shows
Yes, MAL genres. The algorithm only uses those so I need your overall top 3 MAL genres in descending order. So the overall top 3 genres that you like the most. The first being your 1# favorite genre, the second one the 2# favorite genre and the third the 3# favorite genre.
Well i was more asking if it uses the themes they have on their site, because out of the 18 genres action adventure and fantasy will clearly be the ones i chose and whats selected because of the shitty system they have. If they could go deeper and select out of themese that would be cool, but i doubt it would be helpful since they dont use many tags on mal.