Scholarlink

<p>This study employs machine learning (ML) and natural language processing (NLP) techniques to analyze how students interact and communicate during the scientific inquiry process in high school science classes. A lesson utilizing the Acid–Base Chemistry module from ChemCollective’s Virtual Labs was conducted with 205 9th-grade students, resulting in the collection of 18,742 discourse exchanges between students. The collected discourse was classified using a BERT-based supervised learning model and an existing discourse analysis framework. Based on the classification results, individual student characteristics were extracted, and student communication patterns were analyzed using <i>K</i>-means clustering and the elbow method. Ultimately, unsupervised learning was utilized to categorize discourse patterns without relying on the established discourse analysis framework. The results demonstrated that the overall accuracy of the Bidirectional Encoder Representations from Transformers (BERT) model was 79%, with precision, recall, and <i>F</i><sub>1</sub> scores also exhibiting satisfactory performance across categories. <i>K</i>-means clustering enabled the identification of specific interaction types, including question-centered discourse, conceptual discussions, and procedural confirmations. In contrast, unsupervised learning revealed six primary discourse types: planning experimental procedure and role distribution, uncertainty and confusion, color change observations, connecting experimental results to theoretical concepts, temperature measurement, and quantitative data measurement. These findings underscore that the application of ML and NLP techniques transcends mere technical categorization to deliver practical educational insights. Leveraging the uncovered discourse patterns, educators are empowered to customize instruction, provide focused feedback, and develop targeted interventions that foster greater student engagement and enhance scientific reasoning. This study demonstrates that ML and NLP approaches have the potential to not only systematically analyze students’ communication, but also enable the development of more effective teaching strategies and learning environments.</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

Analyzing Student Communication Patterns in Science Classes Using Machine Learning and Natural Language Processing: A Case Study on High School Science Education

  • 期刊论文

  • 发表日期：

    2025年5月2日

查看全文
  • Cheol-Hong Jeon,
  • Jung-Yun Shin,
  • Suna Ryu

摘要

This study employs machine learning (ML) and natural language processing (NLP) techniques to analyze how students interact and communicate during the scientific inquiry process in high school science classes. A lesson utilizing the Acid–Base Chemistry module from ChemCollective’s Virtual Labs was conducted with 205 9th-grade students, resulting in the collection of 18,742 discourse exchanges between students. The collected discourse was classified using a BERT-based supervised learning model and an existing discourse analysis framework. Based on the classification results, individual student characteristics were extracted, and student communication patterns were analyzed using K-means clustering and the elbow method. Ultimately, unsupervised learning was utilized to categorize discourse patterns without relying on the established discourse analysis framework. The results demonstrated that the overall accuracy of the Bidirectional Encoder Representations from Transformers (BERT) model was 79%, with precision, recall, and F1 scores also exhibiting satisfactory performance across categories. K-means clustering enabled the identification of specific interaction types, including question-centered discourse, conceptual discussions, and procedural confirmations. In contrast, unsupervised learning revealed six primary discourse types: planning experimental procedure and role distribution, uncertainty and confusion, color change observations, connecting experimental results to theoretical concepts, temperature measurement, and quantitative data measurement. These findings underscore that the application of ML and NLP techniques transcends mere technical categorization to deliver practical educational insights. Leveraging the uncovered discourse patterns, educators are empowered to customize instruction, provide focused feedback, and develop targeted interventions that foster greater student engagement and enhance scientific reasoning. This study demonstrates that ML and NLP approaches have the potential to not only systematically analyze students’ communication, but also enable the development of more effective teaching strategies and learning environments.