Deep semi-supervised learning for recovering traceability links between issues and commits

DSSLink used different equipment for experiments based on the computing resource requirements of the specific method being enhanced.

For T-BERT, original, self-training and label propagation models are run on a machine equipped with an Nvidia RTX 2080Ti GPU. And TraceFUN is also running on this machine.

For BTLink, original and self-training models are run on a machine equipped with an Nvidia RTX 3090 GPU, while label propagation is run on a machine equipped with an Nvidia A100-PCIE-40GB GPU.

Run the following command to install software dependencies (Python >= 3.8) on a Linux platform.

pip install -r requirements.txt

This repository contains three folders, corresponding to DSSLink enhanced T-BERT, DSSLink enhanced BTLink and TraceFUN.

DSSLink-Enhanced-T-BERT
DSSLink-Enhanced-BTLink
Tracefun

In DSSLink, each folder contains two sub-folders for deep semi-supervised learning, namely self-training and label_propagation.

In each deep semi-supervised learning folder, there are three important .py files. Their functions are introduced as follows:

| File             | Description                   |
| ---------------- |:-----------------------------:|
| run.py           | The entry file                |
| train_stage1.py  | The baseline model            |
| train_deepssl.py | Deep semi-supervised learning |

where the train_stage1.py is used for the training of the original model and the pre-training of DSSLink.

Download the dataset from Google Drive and save it to a suitable location.

First, specify the location of the dataset in run.py.

datasets_dir = 'Your datasets location'

For T-BERT, the models and evaluation results will be saved in outputs_dir.

For BTLink, the models will be saved in output_dir and the results will be saved in result_dir. Please specify them in run.py.

# T-BERT
outputs_dir = 'Your model and results location'
# BTLink
output_dir = 'Your model location'
result_dir = 'Your results location'

• If you need to modify the training parameters of network, such as learning rate and batch size, etc. It can be set specifically through the train_stage1.py and train_deepssl.py.
• After all settings are complete, run run.py to start executing tasks.

• TraceFUN includes a labeling folder for creating new links with unlabeled data, run run.py in the labeling folder to start labeling.

In run.py, the variable frac represents the proportion of the amount of labeled data, and the number of new links to be created is determined by setting the variable frac. The default values of the variable frac for the Flask, Pgcli and Keras projects are 1.1, 1.1 and 0.05 respectively.

• After creating new links, run run.py in the TraceFUN folder to start the TraceFUN task.

• The T-BERT method evaluated in this paper can be found at this link.
• The BTLink method evaluated in this paper can be found at this link.

The pre-trained BERT models used in T-BERT and BTLink can be obtained below:

• RoBERTa-large
• CodeBERT

The number of models we trained is large, which you can obtain them from the following four links:

• DSSLink Trained Models for T-BERT 1
• DSSLink Trained Models for T-BERT 2
• DSSLink Trained Models for BTLink 1
• DSSLink Trained Models for BTLink 2

@article{ZHU2024112109,
title = {Deep semi-supervised learning for recovering traceability links between issues and commits},
journal = {Journal of Systems and Software},
volume = {216},
pages = {112109},
year = {2024},
issn = {0164-1212},
doi = {https://doi.org/10.1016/j.jss.2024.112109},
url = {https://www.sciencedirect.com/science/article/pii/S0164121224001547},
author = {Jianfei Zhu and Guanping Xiao and Zheng Zheng and Yulei Sui},
keywords = {Traceability links recovery, Deep semi-supervised learning, Self-training, Label propagation, Issue and commit}
}