WALGES: Weighted Probability Based Scoring Approach for Solving Algebraic Word Problems using Semantic Parsing

Virtual assistants like Google Assistant, Siri, Cortana, and many others, are now the major feature of smart-phones and tablets. Natural language understanding is the most promising part of those. Besides question and answering, sometimes those smart assistants face verbally stated mathematical problems. Algebraic word problems are one of the fundamental mathematical problems where financial news, election results, sports results, and many other services are also related. In this paper, we have generated solutions of the algebraic word problems. Semantic parsing is used to ground the problem text into containers, quantities, and entities to generate equation trees. From the equation trees, features are extracted to train a local and a global classifier model. After that, to find the candidate equation from the generated equation, probabilistic scores from the local and the global classifier models have used. Backpropagation neural network algorithm is used to incorporate the probabilistic scores of the local and the global classifier models with weights. Our proposed weighted probability based scoring approach increases the accuracy to predict the solution of the algebraic word problems about 9% surpassing the existing system. Our result showed that our proposed technique is useful.