Human-Machine Co-boosted Bug Report Identification with Mutualistic Neural Active Learning

Description This repository accompanies the paper on Human–Machine Bug Report Identification and provides all materials required to reproduce the experimental results. It includes: Source code implementing the MNAL (Model-based Neural Active Learning) approach Scripts for experiments corresponding to RQ1–RQ4 Data preprocessing and model initialization pipelines Result analysis utilities for reproducing tables and figures Additional experimental code for discussion studies Repository Structure .├── data_gen.py # Data preprocessing├── model_gen.py # Model initialization (warm-up training)├── rq1.py # Experiment for RQ1├── rq2.py # Experiment for RQ2├── rq3.py # Experiment for RQ3├── rq4.py # Experiment for RQ4├── rq4_HINT/ # Reproduction of HINT method (ICSE 2024)├── result_analysis.py # Result analysis (tables & figures)├── discussion/ # Additional experiments├── README.md Data Training set:https://tickettagger.blob.core.windows.net/datasets/nlbse23-issue-classification-train.csv.tar.gz Test set:https://tickettagger.blob.core.windows.net/datasets/nlbse23-issue-classification-test.csv.tar.gz Full experimental results:https://www.dropbox.com/scl/fo/o45rrmaolsvnfp8zldqox/h?rlkey=zkqrpev4qqpxyftr9jukvnk45&dl=0 Reproducibility Environment: Python 3.10 PyTorch 1.12.1 CUDA 11.7 Example command: python rq1.py --initial_size <INT> --query_size <INT> --method_setting <METHOD> --start_from_run <RUN> --start_from_step <STEP> Result Analysis Generate tables: python result_analysis.py --table <TABLE_ID> Generate figures: python result_analysis.py --fig <FIG_ID> Additional Experiments The discussion/ directory includes experiments on: Sampling strategies Imbalanced datasets Upper-bound performance Shell scripts are provided for execution. Reference Gao et al., Learning in the Wild: Towards Leveraging Unlabeled Data for Effectively Tuning Pre-trained Code Models, ICSE 2024.