BTLink

This repo provides the code for reproducing the experiments in BTLink : Automatic link recovery between issues and commits based on pre-trained BERT model. Specially, we propose a novel BERT-based method (i.e., BTLink), which applies the pre-trained models to automatically recover the issue-commits links.

Dataset

We build labeled link datasets from open-source software projects to verify the effectiveness of BLink in the within-project and cross-project contexts. You can get the data through the following link (Google drive)：https://drive.google.com/drive/folders/1mjJrscTS63dXt0fwYlqeifg7P1GzmiPU?usp=sharing

Dependency

• pip install torch
• pip install transformers
• pip install sklearn
• pip install nltk

Relevant Pretrained-Models

BTLink mainly relies on the following pre-trained models as the backbone to obtain the embedding representation of NL-NL (issue text-commit text) pairs and NL-PL (issue text-commit code) pairs and obtain feature vectors to complete subsequent link recovery.

• NL-NL Encoder: RoBERTa-large
• NL-PL Encoder: CodeBERT

Besides, you can get our trained model and reproduce the experimental results from the link: BTLink_saved_models.

Start

First, you shold download the dataset from our link.

You can reproduce the results of within-project link recovery by running the file or reproduce the results of cross-project link recovery by running the file.

Please follow the instructions to complete the reproduction：within-project link recovery, Cross-project link recovery.

Result

We present the average performance of BTLink on the within-project link recovery task and the cross-project link recovery task. You can find more detailed results under the results folder.

Within-project(%)

Model	F1	Recall	Precision	MCC	AUC	ACC	PF
BTLink	80.03	84.34	78.81	77.53	90.22	94.76	3.91
FRLink	55.89	49.39	66.58	50.48	72.28	89.43	4.82
DeepLink	38.50	66.47	29.43	27.95	70.21	73.22	26.05
hybrid-linker	38.04	30.63	72.03	36.46	62.75	88.02	5.12

Cross-project(%)

Model	F1	Recall	Precision	MCC	AUC	ACC	PF
BTLink	62.19	71.02	62.72	59.06	81.65	89.75	7.73
FRLink	49.32	48.80	62.73	45.83	71.49	87.92	5.81
DeepLink	13.92	53.12	17.95	6.58	54.89	56.30	43.35

Reference

If you use this code or BTLink, please consider citing us.