build discovery engines instead of search. Search minimizes diversity of result, discovery maximizes diversity under similarity constraints. This is a much more interesting problem, what similarity and diversity constraints are appropriate.
as an example of the limitations of search, searching for a document by typing the whole thing will give you that same document. Discovery engines would likely not be short queries, but entire documents.
Do you have an example of such an engine?
e.g. pubmed recommends related articles.
I haven’t seen anything net-wide, tho.
I would say that, although problematic, the recomendation system for youtube is an example of a Discovery Engine.