The problem is that they are "black box" collaborative filtering systems, without transparency and without giving the individual user the ability to control or configure them.

I suspect that a sufficiently flexible toolbox allowing one to configure and teach one's own recommendation system might do the job.

I am not optimistic about how far a "black box" collaborative filtering system might go (they are great for vendors, because they make results for an average customer somewhat less mediocre, and this translates to profit, but for an individual user results still remain mediocre, especially for the outlier users with tastes far from the average.)