Scholarlink

<p>The essence of effectively tackling geometry math word problems (MWPs) lies in extracting information from question text using natural language processing (NLP) and diagram interpretation using Image Processing. This enables the precise application of mathematical axioms and theorems, transforming complex questions into solvable equations. The primary focus of this work is on the text parsing step, which is difficult to accomplish with current approaches since they rely on handwritten logic forms or regular expressions, neither of which are very flexible or scalable. In this work, we leverage the seq2seq with attention based encoder-decoder model and improve it by adding Named Entity Recognition (NER) tags. Additionally, the proposed method in this work incorporates an approach to identify and generalize numbers within the text called as generalized variable approach in the literature. We conduct a comparative analysis against ground-truth logic forms meticulously created by domain experts to evaluate the precision of our generated logic forms. Furthermore, we assess our findings with Jaro-Winkler similarity scores on a custom dataset with 500+ questions and the publicly available dataset Geometry3K, which yield scores of 0.69 and 0.81, respectively. We also evaluate the proposed system using the METEOR (Metric for Evaluation of Translation with Explicit Ordering) metric, scoring 0.67 and 0.59 on the custom and Geometry3K datasets, respectively. Furthermore, we evaluate the integration of logic forms generated from text with those extracted from diagrams, demonstrating improved structural completeness and overall accuracy in the resulting representations.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Extraction of structured logic forms from geometry problem descriptions

  • 期刊论文

  • 发表日期：

    2026年5月21日

查看全文
  • Archana Boob,
  • Tarun Landa,
  • Dweeja Reddy Devi,
  • Sushant Kogurwar,
  • Divya Durga Kanuparthi,
  • Mansi Radke

摘要

The essence of effectively tackling geometry math word problems (MWPs) lies in extracting information from question text using natural language processing (NLP) and diagram interpretation using Image Processing. This enables the precise application of mathematical axioms and theorems, transforming complex questions into solvable equations. The primary focus of this work is on the text parsing step, which is difficult to accomplish with current approaches since they rely on handwritten logic forms or regular expressions, neither of which are very flexible or scalable. In this work, we leverage the seq2seq with attention based encoder-decoder model and improve it by adding Named Entity Recognition (NER) tags. Additionally, the proposed method in this work incorporates an approach to identify and generalize numbers within the text called as generalized variable approach in the literature. We conduct a comparative analysis against ground-truth logic forms meticulously created by domain experts to evaluate the precision of our generated logic forms. Furthermore, we assess our findings with Jaro-Winkler similarity scores on a custom dataset with 500+ questions and the publicly available dataset Geometry3K, which yield scores of 0.69 and 0.81, respectively. We also evaluate the proposed system using the METEOR (Metric for Evaluation of Translation with Explicit Ordering) metric, scoring 0.67 and 0.59 on the custom and Geometry3K datasets, respectively. Furthermore, we evaluate the integration of logic forms generated from text with those extracted from diagrams, demonstrating improved structural completeness and overall accuracy in the resulting representations.