It doesn't learn, it fully supervised and doesn't have feedback loop.
I don't say it's not interesting, it's step forward compared to other articles in my collection [1][2][3] about the same subject allows to dive easily into the metrics and heuristics used in IR.