CCBERT

Code for paper(accepted by Journal of Information and Software Technology) "Improving Stack Overflow question title generation with copying enhanced CodeBERT model and bi-modal information"

CCBERT is a Copying enhanced CodeBERT model for Stack Overflow title generation.

Package Requirement

To run this code, some packages are needed as follows: Python version 3.7

OpenNMT-py == 2.0.1
pytorch == 1.6.0
rouge == 1.0.0
nltk == 3.6.2
transformers == 4.6.1

Dataset and Models

Here is the directory structure of our repository:

├─data_prepare
└─models
    ├─ccbert
    ├─elastic
    ├─onmt_lstm
    └─oracle

data_prepare

Files in this directory are the scripts for data pre- and post-processing, including all the python code for filtering, cleaning, partitioning, counting and figure drawing.

Our dataset are composed of two parts: the Python subset and the Java subset. We upload those processed data to kaggle.com for intuitive interaction. Stack Overflow Title Generation (py&java filtered) | Kaggle

models

We put the code for CCBERT and all baseline models in this directory except BART, for which we use fairseq's implementation.

Specifically, to run CCBERT, one should first view and change the basic configures like file_path, version_id in the config.py. Then run the main.py for training and evaluation.

Citation

If you find this work inspiring for your research, please cite our paper:

@misc{zhang2021improving,
      title={Improving Stack Overflow question title generation with copying enhanced CodeBERT model and bi-modal information}, 
      author={Fengji Zhang and Jacky Keung and Xiao Yu and Zhiwen Xie and Zhen Yang and Caoyuan Ma and Zhimin Zhang},
      year={2021},
      eprint={2109.13073},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}