Graph Constraints Refined for Transitive Relations

Luhan Zhen,Yonggang Zhang,Zhanshan Li

DOI: https://doi.org/10.1016/j.knosys.2023.110458

IF: 8.139

2023-01-01

Knowledge-Based Systems

Abstract:As a powerful modelling tool, Constraint Satisfaction Problems (CSPs) can be solved efficiently using various solving paradigms and can encapsulate various types of constraints in a single model. Graph constraints are one such type of constraint. Existing graph constraints are dedicated to solve classical problems, such as constrained path problems and subgraph extraction problems. However, in some scenarios modelled with CSPs, there are relations where a formal path is formed by selecting a set of vertices on a graph, and the shortest path between two neighbouring vertices on that path needs to be restricted. Such relations are known as transitive relations on the graph, which cannot be expressed by existing graph constraints. Although expressive table constraints can be used to express transitive relations, this approach significantly reduces the performance of using table constraints. Therefore, to express transitive relations on a graph that can be efficiently modelled and reasoned by various types of CSP solvers, we propose two new constraints and present propagation algorithms based on generalized arc consistency. Finally, we perform comparative experiments to demonstrate the correctness and efficiency of the proposed approach. The results of the experiments not only prove the correctness of the propagation algorithms but also show that our proposed approach of modelling directly using new constraints and then reasoning using proprietary propagation algorithms, is more efficient than translating relations into table constraints and then reasoning.