A method and system for efficiently answering a local neighborhood aggregation query over graph data. A graph which has a plurality of nodes is received and stored in memory. A local neighborhood aggregation query is received. A processing engine applies forward processing with differential index-based pruning, backward processing using partial distribution, or an enhanced backward processing that combines the backward processing and the forward processing. As a result of the forward, backward, or enhanced backward processing, nodes in the graph that have the top-k highest aggregate values over neighbors within h-hops of the nodes are determined. Identities of entities or persons associated with the determined nodes are presented and/or stored.