Learning to Reason Deductively: Math Word Problem Solving as Complex Relation Extraction

Jie, Zhanming; Jierui, Li,; Lü, Wei

doi:10.18653/v1/2022.acl-long.410

Cited by 32 publications

(28 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Following Jie et al (2022), we also adapt a rationalizer to update the representations of all quantities and intermediate expressions at the end of each step. This module is crucial because if the representations are not updated, those expressions that were initially highly ranked would always be preferred.…”

Section: Rationalizermentioning

confidence: 99%

“…RoBERTa-GTS and RoBERTa-Graph2Tree replace the original encoder of GTS and Graph2Tree with RoBERTa. The most similar work of our approach is RoBERTa-DeductiveReasoner (Jie et al, 2022), which uses RoBERTa as the encoder and decodes the math- ematical expression using a bottom-up approach. However, it encodes the problem text as a whole without any special attention to the question text and fails to preserve any mathematical law when computing the representation for each candidate expression.…”

Section: Baselinesmentioning

confidence: 99%

“…All experiments are run 3 times for each dataset on a cluster with 4 NVIDIA RTX-8000 GPUs. Following previous works (Lan et al, 2022;Jie et al, 2022), we report the 5-fold cross-validation accuracy on MAWPS and the test accuracy on Math23k, MathQA, and SVAMP. The results of the baseline approaches are taken from their original papers.…”

Section: Implementation Detailsmentioning

confidence: 99%

“…To analyze our techniques in detail, we further perform an ablation study to explore the effectiveness of our proposed two techniques. We replace our constructor with the module in RoBERTa-DeductReasoner (Jie et al, 2022) without preserving the step-wise communitative law, our complete model improves the test accuracies by 0.3%, 1.2%, 0.2% on MAWPS, MathQA, and SVAMP respectively. Exceptionally, the accuracy decreases by 0.5% on Math23k.…”

Section: Ablation Studymentioning

confidence: 99%

“…They use an encoder to embed the problem text using a language model (e.g., GRU (Chung et al, 2014), BERT (Devlin et al, 2019)) and then employ a decoder to generate an expression sequentially. Later works (Xie and Sun, 2019;Zhang et al, 2020b;Cao et al, 2021;Jie et al, 2022) develop Seq2Tree or Seq2DAG neural models, where the decoders produce operators and quantities based on a specific structure iteratively. Despite the recent progress, these neural solvers are still miles away from building robust representations and effective solutions like humans.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Deep Learning in Automatic Math Word Problem Solvers

Zhang

2022

AI in Learning: Designing the Future

View full text Add to dashboard Cite

The design of an automatic solver for mathematical word problems (MWPs) dates back to the early 1960s and regained booming attention in recent years, owing to revolutionary advances in deep learning. Its objective is to parse the human-readable word problems into machine-understandable logical expressions. The problem is challenging due to the existence of a substantial semantic gap. To a certain extent, MWPs have been recognized as good test beds to evaluate the intelligence level of agents in terms of natural language understanding and automatic reasoning. The successful solving of MWPs can benefit online tutoring and constitute a milestone toward general AI. In this chapter, we present a general introduction to the technical evolution trend for MWP solvers in recent decades and pay particular attention to recent advancement with deep learning models. We also report their performances on public benchmark datasets, which can update readers’ understandings of the latest status of automatic math problem solvers.

show abstract

Section: Rationalizermentioning

confidence: 99%

Section: Baselinesmentioning

confidence: 99%

Section: Implementation Detailsmentioning

confidence: 99%

Section: Ablation Studymentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Deep Learning in Automatic Math Word Problem Solvers

Zhang

2022

AI in Learning: Designing the Future

View full text Add to dashboard Cite

show abstract

Does a language model “understand” high school math? A survey of deep learning based word problem solvers

Sundaram,

Gurajada,

Padmanabhan

et al. 2024

WIREs Data Min & Knowl

View full text Add to dashboard Cite

From the latter half of the last decade, there has been a growing interest in developing algorithms for automatically solving mathematical word problems (MWP). It is a challenging and unique task that demands blending surface level text pattern recognition with mathematical reasoning. In spite of extensive research, we still have a lot to explore for building robust representations of elementary math word problems and effective solutions for the general task. In this paper, we critically examine the various models that have been developed for solving word problems, their pros and cons and the challenges ahead. In the last 2 years, a lot of deep learning models have recorded competing results on benchmark datasets, making a critical and conceptual analysis of literature highly useful at this juncture. We take a step back and analyze why, in spite of this abundance in scholarly interest, the predominantly used experiment and dataset designs continue to be a stumbling block. From the vantage point of having analyzed the literature closely, we also endeavor to provide a road‐map for future math word problem research.This article is categorized under: Technologies > Machine Learning Technologies > Artificial Intelligence Fundamental Concepts of Data and Knowledge > Knowledge Representation

show abstract