Paper Title
Analysis of Linguistics and Math Features for Classification of Math Word Problems: Insights and Future Direction
Abstract
Having math word problems (MWP)of varying difficulty levels can help instructors in identifying the knowledge
levels of learners in teaching-learning systems. Given a large database of MWPs instructors spend significant time
customizing content to meet learner needs. In this paper,Machine learning (ML) and AI-based methods are proposed to
automatically classify math word problems. MWPs involve mathematical equations, symbols, and operators in addition to
linguistic complexities. This paper presents various challengesin identifying and extracting relevant linguistics features as
well as mathematical features that can aid in the automatic classification of MWPs. Based on our study we found that there is
improvement in F1-score for a 3-level difficulty when compared to 5-level difficulty label of MWPs. Our study underscores
the importance of further enhancing the feature set and developing appropriate mathematical tokenizersto improve the model
performance.
Keywords - Math Word Problem (MWP), Difficulty Level of MWP, Adaptive Learning Systems (ALS)