TfIdf still state of the art?

[asmaier]

This is more a question:
From looking at the benchmark results

https://github.com/boudinfl/pke/blob/master/results.md

it seems simple TfIdf outperforms every other algorithm on the inspec dataset
not only in speed, but also in f-score.
Is that really the correct conclusion? Or am I misunderstanding the results?
Is TfIdf really state of the art for keyphrase extraction?