Enhanced Graph-Based Model with Mathematical Knowledge Embedding for Math Word Problem Solving

Math Word Problems (MWPs) represent a critical area of research in natural language processing and artificial intelligence, with the goal of translating natural language descriptions into mathematical expressions and deriving solutions. Despite recent progress with graph-based and tree-based neural models, many approaches still struggle to capture the implicit mathematical knowledge embedded in these problems. This limitation impedes their ability to accurately represent the problem-solving process, especially for complex MWPs that require an understanding of implicit mathematical relationships. To tackle this challenge, we propose integrating mathematical knowledge directly into graph neural network models to enhance their effectiveness in solving intricate MWPs. Specifically, we introduce a formula graph where each formula is abstracted into a graph structure and integrated into an enhanced model, thereby facilitating efficient resolution of MWPs involving implicit knowledge. Our approach aims to improve model performance on complex MWPs, as demonstrated through empirical evaluation on the Ape210K dataset.