After several experiments, the MRR metric for the JF17K dataset in the paper was changed to "0.580". Note that this does not affect the experimental result that HyConvE is still optimal.