I’ve also found that recommendation systems are of much use actually when you are looking trough some things or areas you are not familiar with since that you have reference points of navigation, trough the materia, or possible things that can be connected and whose connections perhaps you wouldn’t be able to see, in which case you are using other peoples experience directly, but on the other hand I can see how this can also limit your movement in a certain way or even become anoying if the system is severely wrong.
The best reccommendations of course I’ve found to be the ones that are directly derived from the point of my interest, if we take for example certain topic I am researching or certain article, in which case recommendations based on the actual reference list for that article, somehow end up being the most precious for me, which then somehow excludes the whole algorithm tactic of recommendations, returning it practically to analog indexing principle.
The thing that is for sure universal for all recommenadation systems is tat it doesn’t actually allow individual approach to the subject, trying to generalize interests, and possible directions based on the information derived from the mase, practically.
As an orientation points, in unknown field I think it works very well but in a field with much more specificity, or precision there is not much weight to it actually.
I’ve found that recommendation systems work the best when combined with couple of tehniqes of collecting data , from the side of the user as well as from the side of the.
For example when I am looking for certain categories of experimental videos on Vimeo, recommenations based on the most popular videos seen align with the one I was already looking for , doesn’t actually give ma good enough source, because there is a big possibility that I will find a video which doesn’t have that big ratings from the side of the users, or viewers, which can in a certain case indicate totally opposite, that that video is not easy to look at or like with thumbs up which comes as a positive information in this kind of search, when I guess searching trough taggs self-referenced from the side of author are the most precious data sources.