@K2 I could answer ”no, I wouldn't”, because they are pretty tricky and usually fail.
But I could also recommend you Bats of The Republic by Zachary Thomas Dodson. It has a plot, but it also employs several different typographical tricks. The story takes place in the past and in a future, and there's some kind of dystopic future which doesn't employ fire or electricity.
@moormaan Number of peers to take into account is good. The book universe is so huge that it's almost impossible to find the pearls without a huge peer group.
There are other aspects than star rating that can be generalised to algorithmic form, no need to limit the engine to that.
And no, the input is not static: 1) I read more all the time, 2) more books are published all the time, 3) recommendations could be made from different subsets at different times.
@smays So far I've found only two kinds of recommendations from LibraryThing:
1) based on the whole library
2) based on individual books.
Is it possible to create a custom collection (a subset of library, thus a middle form between the two mentioned above) and get recommendations based on that?