Natural Language Processing based Math Word Problem Solver and Synoptic Generator

Math word problem solution generation is a process where the machine itself will automatically generate/provide a final answer as well as a step-wise solution for the provided Math word Problem. This simply measures the machine's capacity and capability to comprehend and reason in natural language. This paper proposes a solution where various techniques are implemented, different models are tried and tested, and the scope is extended to cover multiple types of Math Word Problems. The domains worked on are speed, distance, time, profit and loss. The main purpose of this paper is to create a system with various domains in a hybrid framework which can accurately produce a step-wise solution for the given Math Word Problem. Based on the different tested approaches, Random Forest (used for Domain Classification) + 2 custom Transformer Model (used for Equation and Formula Generation) are found to be the most suitable method which provided the best accuracy.

I. Introduction

From the standpoint of Natural Language Processing (NLP), we need to transform the natural language text to mathematical formulas that can be instantaneously solved. Currently, web browsers are incapable of solving mathematical word problems; if a viewer queries the answer to a math problem, traditional web browsers will only return the calculated result. Here, viewers' learning abilities are harmed when only the result is returned because they are not learning on how to address the problem. The purpose of this paper is to solve the Math Word Problem that has been addressed and provide a stepwise analytical solution that gives more insight about the word problem and lets the user grasp and acknowledge the adequate solution. It can help teachers, students and online bot assisted forums.