Forums

Genres are implicitly understood by the algorithm. Basically, the system understands genres a little different than what one would expect. It doesn't take MAL genres at all, it just tries to determine things based on what OTHER PEOPLE like. See the about page if you want a general explanation.

Again, if you give low ratings to old anime, it will not recommend them. Time-based filtering is not necessary because if you correct the recommendations by giving low ratings to old anime, the algorithm will implicitly understand that you don't like old anime.

Recommendation engines only work well when it knows what you like and dislike. If it has no indication that you dislike old anime, then it will not produce good results. Recommendation engines work best on multiple train-recommend cycles -- if the recommendations are wrong, you can retrain the results against the areas where it is wrong. There's a difference between recommendation engines and being psychic.

It is giving you those recommendations because people that tend to have similar anime ratings as you like those anime. If you're a weird case, it may take some correction cycles before it gives you correct results -- for most people the first recommendation it gives will give good results with no corrections.

Another recommendatory system https://animeadvice.me/ . Works the same as animecf.

Very nice. Thanks for sharing it.

This thread is Hall of Fame worthy. Hopefully this can be stickied in the relevant areas.