Entity-Dependency Graph Enforced Quantity Relation Extraction for Solving Arithmetic Word Problems

An entity-dependency graph (EDG) is proposed to efficiently represent mathematical and non-mathematical relations of arithmetic word problems. The EDG of a problem contains nodes representing noun entities and links representing entity-dependencies. An algorithm was proposed to generate EDG which is used to identify entity associated with quantity relation in arithmetic word problems. The results are used to extract quantity relations. Through the experiments, we demonstrate that this method can improve the efficiency of quantity relation extraction by syntax-semantic method. The EDG can also be used to improve the interpretability of the problem solving.