講演情報
[7A-02]Expression based Math Word Problem Generation Via Semi-Automated Method
*Shin Dongha1、Iwaihara Mizuho1 (1. 早稲田大学 IPS Data Engineering LAB)
発表者区分：学生
論文種別：ショートペーパー
インタラクティブ発表：あり
論文種別：ショートペーパー
インタラクティブ発表：あり
キーワード：
Math Word Problem Solving、Arithmetic Reasoning、Natural Language Generation、Large Language Model
Mathematical Word Problem (MWP) is one of the subtask in reasoning. It presents mathematical exercises as narratives, requiring translation of textual descriptions into mathematical expressions. Unlike standard mathematical problems, MWP challenge solvers to interpret and extract relevant information from the narrative. This process enables the evaluation of solvers across multiple skills, including reading comprehension, information identification, text-to-mathematics translation, and mathematical problem-solving. The Chain-of-Thought (CoT) reasoning framework has recently catalyzed significant advancements in the reasoning capabilities of large language models (LLMs), leading to notable performance improvements. The MWP is also the one of the key area where this framework has demonstrated success. However, most existing MWP datasets are primarily focused on linguistic features, such as lexical diversity and grammatical precision, which limits their scope in terms of mathematical complexity and the diversity of equations presented. Consequently, studying the full potential of LLMs in tackling a broader array of mathematical concepts remains underexplored. To bridge this gap, we propose the generation of a MWP dataset using GPT-4, incorporating a wider range of mathematical formulas and topics. In this study, mathematical problems were generated using GPT-4o, followed by a manual verification process conducted by the researcher to ensure accuracy and quality. This approach enabled the creation of approximately 978 problems spanning around 13 diverse mathematical topics.