I completely disagree with the proposed solutions to this problem. The way to solve this is to:

1. Have enough users that the effect of differing star rating meanings doesn't matter
2. Have a way to view your friends ratings separately to the average rating

The reason why you can't have a NormalizedRating is because this is not a concept in reality. We can't know we see the same colour green nevermind know that PersonA's 5* is the same as PersonB's 3*, it's just not a useful use of effort

I know this is a bit old of a request but I'm wondering if what you're talking about is the same thing that mangadex and myanimelist have for their ratings? I would actually like that, actually. I also don't really think you need ai for that.

Like their 7 star rating means "fine". My fine (negative) might be different from your fine (positive) but it's still "fine". But then there's the half star thing to think about.

Star ratings are borked and of very little value, so why not go for something completely different?

Reviews, especially from the people you follow and know their tastes and preferences of, could be a much more valuable indication of whether the book is for you.

An AI approach based on your own preferences, the "properties" of the book and how well the book has fared with your close friends could potentially give much more useful recommendations than star ratings.

This is something we'd like to do when we generate the average rating for each book.

Or in the meantime something that defines each rating level?