Our study presents the BET-BiLSTM model, a robust framework for automated requirements classification. By utilizing ensemble features from transformer models (BERT, RoBERTa, XLNet, GPT-2, T5) and applying advanced data balancing, we achieve improved classification accuracy and reliability. The model efficiently predicts labels for missing and unlabeled requirements, outperforming conventional methods with its practical application across a variety of datasets.

This paper evaluates if the existing traceability approaches can accurately identify test-to-code links in Python projects. The results reveal that the performance of test-to-code traceability approaches on Python has many differences with Java, including (1) most of the existing techniques have poor effectiveness for Python; (2) after augmenting with cross-level information, the recall significantly drops; (3) machine learning based combination approach achieves the best recall but the worst precision.

In this paper, a multilabel emotion classification method is proposed for software engineering text by utilizing TextCNN, word embedding, and hyper-parameter optimization. The method shows superior performance compared with previous approaches, achieving F1-Micro scores of 84.6001% and 76.9366% on the Jira and Stack Overflow datasets, improving by 3.5001% and 8.6366%, respectively. These results indicate significant improvement in the classification of emotions within software developers' communication channels, potentially leading to more effective collaboration and increased productivity.