AlexanderColen/Annie-May-Discord-Bot

Use mean score instead of count when determining tendencies.

Closed this issue · 3 comments

AlexanderColen commented

Score will actually show taste, whereas count will just show the most & least common genres.

oridistor commented

Hi, can I offer an idea? I think I might actually try to implement it and send you a pr.

I think about maybe making this configurable with the guild setup, so you can decide how you want to rank things in your server.

There will be 4 possible metrics:

  1. count (as it is right now)
  2. mean_score (like I offered originally)
  3. total time watched (another sortable metric that anilist offers)
  4. Weighted mean_score.

Weighted mean score will be marked for being experimental and a bit more process heavy, and will calculate for each item the score personally by weighing in the numbers of the item to a certain level, according to how many items you watched.

Like I mentioned, I have no problem with implementing such an algorithm or even trying to do the "configurable weighting"

AlexanderColen commented

The weighted mean score actually sounds like the best idea if it is implemented properly.

Feel free to shoot in a PR for the weighted mean_score, for now I will change the default to mean_score and look into making it customizable per server.

AlexanderColen commented

Hi, can I offer an idea? I think I might actually try to implement it and send you a pr.

I think about maybe making this configurable with the guild setup, so you can decide how you want to rank things in your server.

There will be 4 possible metrics:

1. count (as it is right now)

2. mean_score (like I offered originally)

3. total time watched (another sortable metric that anilist offers)

4. Weighted mean_score.

Weighted mean score will be marked for being experimental and a bit more process heavy, and will calculate for each item the score personally by weighing in the numbers of the item to a certain level, according to how many items you watched.

Like I mentioned, I have no problem with implementing such an algorithm or even trying to do the "configurable weighting"

Instead of keeping track of it in here, I made a separate issue for it. (#98) Please redirect the updates on potential PR progress there.

I changed it to use meanScore by default for now and added some extra tendencies to look at minutesWatched and chaptersRead as well.