Supplementary Material of "NoRBERT: Transfer Learning for Requirements Classification"
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
This is the supplementary material of the paper "NoRBERT: Transfer Learning for Requirements Classification" at RE20. In this paper we explore the performance of transfer learning (with Google's language model BERT) on different tasks in requirements classification. Especially the performance on projects, completely unseen during training, is in the focus of the paper.<br> Additionally, we developed a new dataset based on the Promise NFR dataset, that includes a more fine-grained labeling of functional requirement based on their concerns (Function, Data, Behavior). This repository contains the datasets and code used in the paper, as well as additional results: Dataset contains the labeled dataset for the classification of functional requirements concerns (based on Promise NFR dataset) as well as information about our labeling (results of each annotator and Krippendorf's Alpha, KALPHA) Code contains the python notebooks (code) and datasets used for Task 1: Binary F/NFR classification (on Promise NFR dataset) Task 2: Classification of most frequent NFR subclasses (on Promise NFR dataset) Task 3: Classification of all NFR subclasses (on Promise NFR dataset) Task 4: Functional and Quality aspects classification (on relabeled Promise NFR dataset) Task 5: Classification of functional requirement concerns (on functional concerns dataset) Notebooks to apply pretrained models for each task to an input requirement and pretrained models for each task Results contains the results of all tested hyperparameter configurations for each task Note that we are not able to provide the actual models that were used to produce the results of the paper.<br> We used cross validation experiments that would result in a huge amount of model files per experiment run on each task.<br> As the model files are quite large this is not feasible.<br> The results may still be reproduced with the supplied notebooks. <strong>Attribution (of datasets used):</strong> The Promise Dataset can be attributed to Jane Cleland-Huang and was provided for the RE'17 Data Challenge.<br> Jane Cleland-Huang, Sepideh Mazrouee, Huang Liguo, & Dan Port. (2007). nfr [Data set]. Zenodo. Available: http://doi.org/10.5281/zenodo.268542<br> RE'17 Data Challenge: http://ctp.di.fct.unl.pt/RE2017/pages/submission/data_papers/<br> See also: Sayyad Shirabad, J. and Menzies, T.J. (2005) The PROMISE Repository of Software Engineering Databases. School of Information Technology and Engineering, University of Ottawa, Canada. Available: http://promise.site.uottawa.ca/SERepository The relabeled dataset can be attributed to Dalpiaz et al: F. Dalpiaz, D. Dell’Anna, F. B. Aydemir, and S. Çevikol, “explainable-re/re-2019-materials,” Jul.2019. https://doi.org/10.5281/zenodo.3309669
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.005 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it