Prompt-Based Missing Entity Recovery for Solving Arithmetic Word Problems

Most existing neural models solve arithmetic word problems from explicit problem text. However, they can hardly give the solution procedure for problems that contain implicit quantity relations. This paper proposes a missing entity recovery(MER) model to solve arithmetic word problems(AWPs) with implicit knowledge. Given an AWP, the model effectively identifies and represents its explicit expressions into the Nodes Dependency Graph(NDG). Then the nodes on the graph get implicit knowledge from the knowledge base in a recursive way. The group of selected nodes is finally transformed into a group of equations using the solving engine to obtain the answers. The proposed algorithm is evaluated practically based on a collection of established datasets Math23K, showcasing its high accuracy in problem-solving and application in various application situations.