Autosorting 'Quality' Indicator trained on 11k+ works. Very generous with small fics, rewards engagement over popularity (bookmarks-collections-comments/kudos instead of hits) with a 0-100 score spread. Sort & position toggles included.
Thanks! I pushed the new update too quickly, and it did produce too many 100s. In fact, the score function had been wrong this whole time; I fixed it and added the new formula in the description along with some graphs.
Max creates a triangular distribution by picking the top metric while Average makes things... average. Extreme 0–100 scores can only happen if all metrics align (all high or all low). This eliminates most high and low scores and brings everything toward the middle.
The new settings should be good to use this time:
The only recommended changes now would be:
Anyway, it's indeed impossible to mix multiple uniform metrics into a uniform score, my previous approach was flawed. For example, it would take an 80 score down to 60 by normalizing after the other metric's 80+ scores were shifted above. The new Max lets it stay at 80 even if it creates more high scores.
But in my experience, after reading all the works with top scores; seeing low scores discouraged me from trying more works unfairly. The Max distribution actually gives me more high scores to enjoy.
I did some comparison with liked and disliked works list - it feels like some kind of sorcery, practically all the works I've put in disliked list have mediocre score at best, it really represents the quality of the work itself. Really appreciate all the effort you put into fine-tuning this model!
I didn't really have many reference points, so that's great to hear! Thanks!
Wow! Just tried it out and its awesome!
By the way, what is your personal rec for the MAX_SCORE value? For some fics, the score changes drastically like 100 to 51, as far as I see AVG mode negates the dominance of one parameter in favor of balance. Perhaps you have already tested in practice which approach better reflects quality, or is it still subjective?
Thank you so much, that's a lot of work!