Scholarlink

<p>In the evolving landscape of software development, effective bug tracking is critical to maintaining high-quality software systems. Manual triaging of bug reports is time-consuming and often leads to redundant work due to duplicate reports. This paper proposes the Automatic Duplicate and Learning-based Model (ADLM model)—an integrated framework that combines Conditional Random Fields (CRF), Long Short-Term Memory (LSTM), and Dragonfly Optimization (DO) to automate bug detection and accurately classify duplicate reports. The model incorporates a structured three-stage pipeline: pre-processing (including tokenization, normalization, and stop word removal), multidimensional feature extraction (textual, temporal, categorical, and contextual), and advanced classification. LSTM captures deep semantic relationships, while CRF models label dependencies for precise classification. Dragonfly Optimization further refines classification accuracy by fine-tuning hyperparameters. Experimental results demonstrate that the ADLM model outperforms existing techniques in terms of accuracy, precision, recall, F1-score, and kappa statistics. This research not only enhances automated bug report handling but also significantly reduces developer workload, thereby improving overall software quality and development efficiency.</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

CRF_LSTM_DO: automated software bug detection deep learning framework

  • 期刊论文

  • 发表日期：

    2025年10月25日

查看全文
  • Shallu Juneja,
  • Gurjit Singh Bhathal,
  • Brahmaleen Kaur Sidhu

摘要

In the evolving landscape of software development, effective bug tracking is critical to maintaining high-quality software systems. Manual triaging of bug reports is time-consuming and often leads to redundant work due to duplicate reports. This paper proposes the Automatic Duplicate and Learning-based Model (ADLM model)—an integrated framework that combines Conditional Random Fields (CRF), Long Short-Term Memory (LSTM), and Dragonfly Optimization (DO) to automate bug detection and accurately classify duplicate reports. The model incorporates a structured three-stage pipeline: pre-processing (including tokenization, normalization, and stop word removal), multidimensional feature extraction (textual, temporal, categorical, and contextual), and advanced classification. LSTM captures deep semantic relationships, while CRF models label dependencies for precise classification. Dragonfly Optimization further refines classification accuracy by fine-tuning hyperparameters. Experimental results demonstrate that the ADLM model outperforms existing techniques in terms of accuracy, precision, recall, F1-score, and kappa statistics. This research not only enhances automated bug report handling but also significantly reduces developer workload, thereby improving overall software quality and development efficiency.