TDS: fast answering reachability queries with hierarchical traversal trees

Daoliang He, Pingpeng Yuan

Abstract: Reachability query is a fundamental operation in graph applications, including social network analysis, knowledge graph based NLP. Current label+G solutions tend to directly answer reachability queries between near vertices. When two vertices are far apart, they have to traverse the graph, which is time consuming. In this work, we propose a multi-dimensional labeling method called TDS to support fast reachability queries. Firstly, we order the vertices with zero-indegree and traverse the graph from them in forward and reverse order to generate two traversal trees. Furthermore, we assign each vertex in a tree two labels, which indicate the level and subgraph it belongs to. With the four labels, many reachability queries for far apart and near pairs can be answered quickly. Experiments on 16 real graphs demonstrate TDS achieves better query performance in a majority of used graphs versus the state-of-the-art, while it consumes near-linear time and space for indexing.