Data-Driven Methods for Solving Algebra Word Problems

TL;DR

This paper evaluates data-driven neural methods for solving algebra word problems, demonstrating that neural classifiers outperform complex models on large datasets, but highlighting the need for semantic and world knowledge for further progress.

Contribution

It shows that well-tuned neural equation classifiers can outperform more complex models on algebra word problems, emphasizing the importance of semantic understanding.

Findings

Neural classifiers outperform sequence-to-sequence models on large datasets.

Error analysis reveals the need for semantic and world knowledge.

Fully data-driven models show promise but have limitations.

Abstract

We explore contemporary, data-driven techniques for solving math word problems over recent large-scale datasets. We show that well-tuned neural equation classifiers can outperform more sophisticated models such as sequence to sequence and self-attention across these datasets. Our error analysis indicates that, while fully data driven models show some promise, semantic and world knowledge is necessary for further advances.