Enhancing Traceability Link Recovery with Unlabeled Data
Run the following command to install software dependencies (Python >= 3.8) on a Windows 10 platform with an Nvidia RTX 2080Ti GPU.
- Tensorflow == 2.6.0
- CUDA == 11.2
- cudNN == 8.1
pip install -r requirements.txt
Before using TraceFUN, standardize software artifact data to a csv file by the following format.
Artifacts
| id | doc |
| ----- | ----- |
| 123 | xxxxx |
Links
| s_id | t_id |
| ----- | ----- |
| 123 | 456 |
TraceFUN uses two sources of data, the CoEST datasets used for training contrastive learning model and the TLR. These two datasets are provided in the folder data.
Run the python file Run_CoEST.py in the folder Data_Process/CoEST_data_process/ to automatically convert the CoEST data to standard format.
Run the python file standardize.py in the folder Data_Process/TLR_data_process/ to automatically convert the data to standard format and divide it into 5 folds of samples for cross-validation.
The standardized data is stored in the folder standard, where should have the following three files:
source.csv # source artifacts
target.csv # target artifacts
link.csv # traceability linksTraceFUN introduces two similarity prediction methods, i.e., vector space model (VSM) and contrastive learning (CL). These two similarity prediction methods are stored in the folder Similarity_Prediction_Method.
Run the script file Run_VSM.sh in the folder Similarity_Prediction_Method/ by Git Bash to get the similarity lists between unlabeled and labeled artifacts by VSM.
Run the python file Training_data_generation_for_CL.py in the folder Similarity_Prediction_Method/ to get the train.csv (Training data for CL).
Run the python file Train_the_model_for_CL.py in the folder Similarity_Prediction_Method/ to train the CL model.
Run the script file Run_CL.sh in the folder Similarity_Prediction_Method/ by Git Bash to get the similarity lists between unlabeled and labeled artifacts by CL.
Run the python file unlabeled_data_labeling.py to get the files of newly labeled links by VSM, CL, and random selection, respectively:
labeling_result_by_vsm.csv
labeling_result_by_cl.csv
labeling_result_by_random.csvThe calculated similarity lists can be found at Similaritylists.
Run the python file new_links2training_set.py in the folder Data_Process/TLR_data_process/ to add the newly labeled links to the training set for TLR evaluation.
The TLR methods (Trace BERT and TraceNN) evaluated in this paper can be found at TraceBERT.
@inproceedings{zhu2022enhancing,
title={Enhancing Traceability Link Recovery with Unlabeled Data},
author={Zhu, Jianfei and Xiao, Guanping and Zheng, Zheng and Sui, Yulei},
booktitle={2022 IEEE 33rd International Symposium on Software Reliability Engineering (ISSRE)},
pages={446--457},
year={2022},
organization={IEEE}
}