A Weakly Supervised Model for Solving Math word Problems

Analysis of Elementary Math Word Problems Based on AI Deep Learning

“…Up till now, there has been a lot of attempts tackling the problem of solving MWPs with supervised deep learning. Researchers differ over the amount of supervision they use in their transformer models, ranging from those advocating very strong supervision or semantic parsing, by outlining the syntactical logic behind the problem and creating corresponding equation templates for the model to fill in [1] to those working with relatively weak supervision, only providing the final numeric answers [2] and improving the model through specific inner means such as looking for possible mistakes and fixing them [3] . Our training method takes the middle ground approach, providing the equations for learning but does not restrict the form of the equations with a template for sake of generalization.…”