Scholarlink

Evaluating student programming assignments is a labor intensive and time consuming activity for evaluators. A programming question can have several approaches to get the desired solution. During evaluation process, it is not feasible to check the functionality of all approaches adopted by the students to verify the results. However, automation of program code evaluation can reduce this burden. The paper proposed a web tool for automatic program code evaluation using transformer-based model CodeBERT and evaluation metric CodeBERTScore. The proposed web tool was tested for six programming languages, namely, C, C++, Java, Java Script, C# and Python, respectively. Out of these programming languages, C++ program code evaluation outperformed with a precision, recall, F1score and F3score of 0.9478, 0.932, 0.9406 and 0.9328, respectively.

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

Automatic Assessment of Program Code Using CodeBERTScore: a Transformer-Based Approach

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Hunny Gaur,
  • Devendra K. Tayal,
  • Amita Jain

摘要

Evaluating student programming assignments is a labor intensive and time consuming activity for evaluators. A programming question can have several approaches to get the desired solution. During evaluation process, it is not feasible to check the functionality of all approaches adopted by the students to verify the results. However, automation of program code evaluation can reduce this burden. The paper proposed a web tool for automatic program code evaluation using transformer-based model CodeBERT and evaluation metric CodeBERTScore. The proposed web tool was tested for six programming languages, namely, C, C++, Java, Java Script, C# and Python, respectively. Out of these programming languages, C++ program code evaluation outperformed with a precision, recall, F1score and F3score of 0.9478, 0.932, 0.9406 and 0.9328, respectively.