Recent studies on event detection (ED) haveshown that the syntactic dependency graph canbe employed in graph convolution neural net-works (GCN) to achieve state-of-the-art per-formance. However, the computation of thehidden vectors in such graph-based models isagnostic to the trigger candidate words, po-tentially leaving irrelevant information for thetrigger candidate for event prediction. In addi-tion, the current models for ED fail to exploitthe overall contextual importance scores of thewords, which can be obtained via the depen-dency tree, to boost the performance. In thisstudy, we propose a novel gating mechanismto filter noisy information in the hidden vec-tors of the GCN models for ED based on theinformation from the trigger candidate. Wealso introduce novel mechanisms to achievethe contextual diversity for the gates and theimportance score consistency for the graphsand models in ED. The experiments show thatthe proposed model achieves state-of-the-artperformance on two ED datasets