Scholarlink

<p>The cold start problem in bug tracking systems arises when there’s insufficient historical data to accurately recommend reviewers for new bugs, leading to delays in bug resolution. To address this, the given paper employs BERTopic, a pre-trained model, to assign new bugs to different software components through a combination of BERT (bi-directional encoder representations from transformers), clustering, and Term Frequency-Inverse Document Frequency (TF-IDF). Thousands of bug reports from Apache projects are processed using the Bug Report Collection System (BRCS) tool. BERT encoding is employed to create embedding using Principal Component Analysis (PCA) for dimensionality reduction and K-means clustering for topic identification. It captures contextual and semantic relationships in the text, enabling effective topic modelling. The resulting topics are then aligned with software project components, forming clusters where developers are ranked by the number of resolved bugs. The performance of the proposed framework is evaluated using Normalized Pointwise Mutual Information (NPMI) score and compared to traditional methods like Latent Dirichlet Allocation (LDA) and Non-Negative Matrix Factorization (NMF). BERTopic outperformed these methods by achieving an average NPMI of 0.726, which shows better topic coherence and interpretability.</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

BERTopic: a model to solve cold start problem in software bug recommendation system

  • 期刊论文

  • 发表日期：

    2025年3月6日

查看全文
  • Vinod Kumar,
  • Anurag Singh Baghel

摘要

The cold start problem in bug tracking systems arises when there’s insufficient historical data to accurately recommend reviewers for new bugs, leading to delays in bug resolution. To address this, the given paper employs BERTopic, a pre-trained model, to assign new bugs to different software components through a combination of BERT (bi-directional encoder representations from transformers), clustering, and Term Frequency-Inverse Document Frequency (TF-IDF). Thousands of bug reports from Apache projects are processed using the Bug Report Collection System (BRCS) tool. BERT encoding is employed to create embedding using Principal Component Analysis (PCA) for dimensionality reduction and K-means clustering for topic identification. It captures contextual and semantic relationships in the text, enabling effective topic modelling. The resulting topics are then aligned with software project components, forming clusters where developers are ranked by the number of resolved bugs. The performance of the proposed framework is evaluated using Normalized Pointwise Mutual Information (NPMI) score and compared to traditional methods like Latent Dirichlet Allocation (LDA) and Non-Negative Matrix Factorization (NMF). BERTopic outperformed these methods by achieving an average NPMI of 0.726, which shows better topic coherence and interpretability.