PyTorch implementation of paper Issue Links Retrieval for New Issues in Issue Tracking Systems accepted by NLDB 2024.

Issue Links Retrieval for New Issues in Issue Tracking Systems

Issue Tracking Systems (ITSs) are used to manage software issues, including feature requests and bug reports. Issues are often interconnected by labelled links to show the relationships between issues. Recent studies in duplicate bug report retrieval (DBRR) aim to retrieve relevant duplicate bug reports o for a query bug report s to form duplicate pairs (s, o). However, the DBRR task has its limitations: (1) only focuses on the issues with the bug type and overlooks other issue types, e.g., task and improvement; (2) only issues with respect to the duplicate link type are retrieved, whereas a query s may have other link types, e.g., relates and depends on, with candidate issues o. This paper goes beyond the DBRR task and proposes a new task of issue link retrieval (ILR) for any issue types and link types. For a given query s, the ILR task will locate all relevant tail issues o with respect to a given link type r to build issue links (s, r, o). This paper presents novel methods using pre-trained language models (e.g., GPT-2) to embed issue links for computing probability scores to retrieve issue links. Our methods are evaluated on four datasets, demonstrating high effectiveness in issue link retrieval.

Install python virtual environment

# install virtual environment module
python3 -m pip install --user virtualenv

# create virtual environment
python3 -m venv env_name
source env_name/bin/activate
# install python packages
pip install requests
pip install -r requirements.txt

Run the code

bash gpu_run_relation_tail_prediction.sh 

or run
python  relation_tail_prediciton.py    --task_name kg    --do_relation_tail_prediction   --data_dir ./data/RedHat  --model_type  BERT   --model_name_or_path  bert-base-cased   --max_seq_length  400   --per_gpu_train_batch_size   16    --learning_rate 5e-5   --gradient_accumulation_steps 2  --eval_batch_size  512   --pre_process_data  ./pre_process_data   --negative   2   --num_train_epochs   2   --output_dir  ./output_RedHat_e2_2_bert_base_max_seq_400/

--do_relation_tail_prediction: set this flag to run the relation tail prediction task.

--data_dir: the path of dataset,

--model_type: name of the model name

--model_name_or_path: the path/type of the pre-trained model.

--max_seq_length: maximum sequence length of input text.

--per_gpu_train_batch_size: batch size for training.

--learning_rate: learning rate for training.

--gradient_accumulation_steps: number of gradient accumulation steps.

--eval_batch_size: batch size for evaluation.

--pre_process_data: the path of pre-processed data.

--negative: number of negative samples per training instance.

--num_train_epochs: the number of training epoches.

--output_dir: the path of output directory.

Data

Download data from here

Contact:

This paper has been accepted by the 29th International Conference on Natural Language & Information Systems (NLDB 2024). The published version can be viewed by this link [](will give). If you use any code from our repo in your paper, pls cite:

@inproceedings{hu2024issue,
  title={Issue Links Retrieval for New Issues in Issue Tracking Systems},
  author={Miao Hu, Zhiwei Lin, Adele Marshall}
    booktitle={The 29th International Conference on Natural Language & Information Systems (NLDB 2024)},
    year={2024}
 }

Feel free to contact MiaoHu ([email protected]), if you have any further questions.