Seems like almost every creative industry (music, video games, art, writing) is having the same issue: Creation & publication tools are getting cheaper & easier to use, which means a lot more people can publish their creative ideas. With such a huge number of choices, discovery is now the issue. IMO discovery of what is truly high quality is still an unsolved problem. Seems like recommendation systems generally just…
Sometimes nobody really tries. It doesn't help that there are a lot of perverse incentive systems out there. I'm approaching this from mostly an Internet-centric perspective:
One observation I've made is that any story I first see by advertising is probably bad, even if I later see it elsewhere - if it were actually any good, I would've seen it in one of the non-advertising-based mechanisms first. But sites have a strong incentive to promote advertisements to the detriment of quality (and the inaccuracy of "hot" lists).
The "zero-initial-following" problem can be solved by showing each story to a random small subset of active readers (since, as big as the supply of crappy stories is, the demand is always higher). This should be smeared across time-of-day, rather than having a "new" queue subject to gamification. There also needs to be a quick "I'm not interested" feedback, with reasons including "breaks site rules", "bad story", "bad grammar", "bad initial hook", "bad continuation", "I just don't like it" (featured prominently), and "this story is badly tagged" (because both positive and negative tag searches should be the primary way of using any reading site).
Some particular ways that tagging implementations can fail:
* categories and tags are different things, thus a tag is often missing
* no tagging for things like "this a fanfiction of", "this is translated from", "author is not a native English speaker", ...
* tag names are ambiguous, meaning completely different things in different contexts
* tag names are contextual, providing a different shade of meaning depending on other tags
* tags are not prominently displayed when actually looking at a work
* user-made tags are permitted, so duplicates and typos are common
* user-made tags are not permitted and essential tags that people wish to search for (or hide) are missing
* hierarchial (DAG, not tree) tags are not supported, thus a tag is often missing (or if present the list takes up too much space)
* no way to specify tag degree (does this just show up in the background, or is it the focus of the work?)
* number of tags is artificially limited to a very small number
* tag is applied but applicable content doesn't appear yet (mostly relevant for when published serially)
Obviously with outright malicious actors, simply fixing these won't fix everything, but they are absolutely needed to function at scale for the honest actors.