Why do we need a secondary score?
The current MAL score is a weighted average score of all people who watched the show, irrespective of their own individual preferences for that particular genre. That means, if someone rated an anime a 10 and another a 3, MAL could give it a rating of 6, which is completely useless to both people who might genuinely enjoy the series (it's rated far below what they would have rated) as well as to the people who would never watch it (it's rated far too high for people who don't normally watch nor enjoy that type of show). Averages are just that. It's also prone to score manipulation, with people creating fake accounts simply to boost or drag down the score for a particular series.
We already had a similar suggestion before
From what I have seen, most suggestions include changing the MAL rating system to allow for decimals (so you could rate an anime a 7.2557527 for example). Another recent thread suggests showing the score breakdown by age and gender. This doesn't solve the problem of not taking into account an individual's taste, which can be wildly different even from people within the same age, gender and ethnic group. And besides, the statistics may not even be all that accurate in the first place.
What is my proposal?
Because the average MAL scores don't take into account individual preferences, my proposal here is to add a secondary score that reflects the taste of each and every individual. That means the secondary score will be different for different people. The idea is to take a weighted average score from people with similar taste to yours. MAL already have a way to measure taste similarity in the form of Affinity. The higher the affinity score, the more similar the anime taste between two users.
In essence, we could use K-nearest neighbors method from machine learning to predict how much an individual would score a given anime, where the distance is measured by the affinity levels.
Conceptually, this makes sense. That would mean if you like the harem genre, the secondary score will not consider scores from people who disliked anime of that type. Or if you like gore and blood, the score will again reflect how much people who also likes gore and blood rated the show.
Again, the idea is not to replace the average MAL score, but instead to supplement it with score relevant to your own personal taste in anime.
But MAL already has a recommendations feature
One major limitation of this feature is that a series that you enjoy might not be in the recommended list, and even if it is, you don't know how much you're likely to enjoy it because there is no scores. An anime page could have 10 'recommended' anime, but you can't tell which ones you will likely enjoy the most. Some of the recommendation can come from people with completely different tastes in anime.
The post is too long, make it short
Add a secondary score showing a prediction of how much someone would rate a show they haven't watched before based on how people with similar tastes (based on affinity level) rated the show. You could call this the 'Affinity score' or 'Predicted Score'.
What is a potential downside?
1) Because this approach relies on having enough people with similar tastes who watched and rated the anime, this might not work if there's not enough people who have rated and share a similar taste to yours. This is likely going to affect new and currently airing anime the most.
2) Another downside is that it may be more difficult to code this secondary score and could take more MAL processing resources.
By the way, K-NN is only a suggestion, there could be even better, smarter ways to implement a secondary, predicted score.
Thoughts? Do you think predicted scores based on affinity level is a good idea? |