Parse, Align and Aggregate: Graph-driven Compositional Reasoning for Video Question Answering. — SciRadar