This is the source code to the paper "Learning Never Stops: Improving Software Vulnerability Type Identification via Incremental Learning". Please refer to the paper for the experimental details.
You can install the required dependency packages for our environment by using the following command: pip install - r requirements.txt.
Simulate our dataset into a continuous stream. Like we split our original dataset into five tasks.
1.Use the py file under data crawling and processing for data processing. Of course, you can directly use the dataset we have processed: Google Drive Link
2.Run VulTypeIL.py. After running, you can retrain the model and obtain results.
3.You can find the implementation code for the RQ1-RQ5 section and the Discussion section experiments in the corresponding folders.
You can obtain our saved model and reproduce our results through the model link
1、cve_with_graph_abstract_commit.json
2、svti.json
3、svti_simple.json
| C/CPP | Java | |
|---|---|---|
| Number of Repositories | 1062 | 362 |
| Number of CVE IDs | 8476 | 775 |
| Number of CWE IDs | 176 | 115 |
| Number of Commits | 9288 | 902 |
| Number of Vul/Non-Vul Function | 17975/335898 | 2433/39516 |
| Success Rate of Graph Generation | 87% | 100% |
Our dataset is licensed under the GPL 3.0, as found in the LICENSE file.