Generating Equations from Math Word Problem Using Deep Learning Approach

Abstract

AbstractBuilding an automatic system to solve Math word problem is very interesting in AI domain. From last four decades Math word problem takes more attention of the researchers. Automatic system of math word problem can increase the effectiveness of e-learning system which is often used in current days. Many research claimed promising result but most of them has worked on small dataset or similar types of problems. It is really difficult to build a system which can work on huge dataset as well as different types of math word problem In this paper, we are trying to build a hybrid deep learning system which can work on large dataset as well as different forms of math word problem. We have done our experiments on Math23K dataset. Our proposed model gives better result than statistical learning and rule based approach. For our experiments we have used Bahdanau attention mechanism which gives better result than traditional methods for math word problem. The accuracy of our proposed mechanism is 0.58 in BLEU score.KeywordsMath word problemDeep learningSeq-to-seq models