A Parallel Implementation of Liveness on Knowledge Graphs under Label Constraints

Knowledge graphs are used extensively in various fields. Labor-intensive and time-consuming error detection for large-scale knowledge graphs significantly increases the need for efficient model checking algorithms in large-scale graphs. Since conventional algorithms are inefficient on such graphs, we propose a new liveness algorithm of knowledge graphs, under a specific given set of labels. By trimming and marking the graphs alternately, a weakly connected components algorithm is then designed to separate graphs into several disjoint sets and a bidirectional forward-backward algorithm computes strongly connected components in each set in parallel. We have implemented our algorithm on a real medical knowledge graph and several open data sets. The evaluation shows that our algorithm achieves up to 11.2x speedup over conventional algorithm with 16 threads.