What does this research mean for the field?

Fine-tuning the BERT language model for requirements classification (NoRBERT) significantly improves generalization to unseen projects, outperforming recent approaches by up to 15 percentage points in F1-score. Novelty: ClaimNovelty.METHODOLOGICAL. Consensus alignment: ConsensusAlignment.NEUTRAL.

August 1, 2020Open Access

NoRBERT: Transfer Learning for Requirements Classification

View Full Paper

Ask AI

Bookmark

To improve the generalization of automated requirements classification across unseen software projects by fine-tuning a pre-trained BERT language model.
Fine-tuned the BERT language model (NoRBERT) for requirements classification tasks across both seen and unseen project settings.
Evaluated the model on the PROMISE NFR dataset and a relabeled version to separate functional and non-functional requirements, as well as non-functional subclasses.
Annotated functional requirements into specific concern categories—function, data, and behavior—and evaluated multi-class classification performance.
Achieved F1-scores of up to 94% on both seen and unseen projects for classifying functional versus non-functional requirements.
Classified the most frequent non-functional requirement subclasses with an average F1-score of 87%, achieving a 15 percentage point improvement over existing approaches in an unseen project setup on the relabeled dataset.
Attained an F1-score of up to 92% when classifying functional requirements by concern (function, data, and behavior).

Discussion

No takes yet. Share an insight, caveat, or question.

Implication

Computational evaluation demonstrates accurate requirements classification across unseen software projects, indicating robust generalization for automated software engineering.

Cite This Study

Hey et al. (2020) studied this question.

synapsesocial.com/papers/6a08b0dc7de338f10b10ec98 https://doi.org/10.1109/re48521.2020.00028

View Full Paper

Ask AI

Bookmark

Implication

Computational evaluation demonstrates accurate requirements classification across unseen software projects, indicating robust generalization for automated software engineering.

NoRBERT: Transfer Learning for Requirements Classification

Key Points

Discussion

Implication

Cite This Study

NoRBERT: Transfer Learning for Requirements Classification

Key Points

Discussion

Implication

Cite This Study

Also Consider

Also Consider