Classifying and Solving Arithmetic Math Word Problems—An Intelligent Math Solver

Abstract

Solving mathematical (math) word problems (MWP) automatically is a challenging research problem in natural language processing, machine learning, and education (learning) technology domains, which has gained momentum in the recent years. Applications of solving varieties of MWPs can increase the efficacy of teaching-learning systems, such as e-learning systems, intelligent tutoring systems, etc., to help improve learning (or teaching) to solve word problems by providing interactive computer support for peer math tutoring. This article is specifically intended to benefit such teaching-learning systems on arithmetic word problems solving by adding an interactive and intelligent word problem solver to assess an individual's learning outcome. This article presents arithmetic mathematical word problems solver (AMWPS), an educational software application for solving arithmetic word problems involving single equation with single operation. This article is based on a combination of a machine learning based (classification) approach and a rule-based approach. We start with classification of arithmetic word problems into four categories (Change, Compare, Combine, and Division-Multiplication) along with their subcategories, followed by the classification of operations (+, -, *, and /) related to different subcategories. Our system processes an input arithmetic word problem, predicts the category and subcategory, predicts the operation, identifies and retrieves the relevant quantities within the problem with respect to answer generation, and formulates and evaluates the mathematical expression to generate the final answer. AMWPS outperformed similar systems on the standard AddSub and SingleOp datasets and produced new state-of-the-art result (94.22% accuracy).