PrAIoritize: Learning to Prioritize Smart Contract Bugs and Vulnerabilities
share
Smart contract vulnerabilities and bugs have become a key concern for software engineers, as they can lead to significant financial losses, reputational damage, and legal issues. Therefore, prioritizing bug fixing for smart contracts is critical to maintaining trust. Due to the lack of tracking tools, prioritizing smart contract-reported bugs is done manually, which is a tedious task, limits bug triaging, and needs specialized knowledge. Towards this end, we propose PrAIoritize; an automated approach for predicting smart contract bug priorities that assist software engineers in prioritizing highly urgent bug reports. PrAIoritize consists of two main phases: 1) automatic labeling, which involves the automatic construction of a smart contract keyword lexicon and the automatic assignment of priority levels to unlabeled bug reports; 2) model construction, which involves feature engineering and designs layers of feed-forward neural networks (FFNNs) and bidirectional long short-term memory (BiLSTM) with multi-class classification to better capture the features of the textual descriptions of bugs and predict their priority levels. The model then is trained using smart contract bug reports collected from two data sources: open-source software (OSS) projects available on GitHub and NVD vulnerability database. Our evaluation demonstrates significant improvement over state-of-the-art baselines and commonly used pre-trained models (e.g. BERT) for similar classification tasks, with 5.75 in F-measure, precision, and recall.
READ FULL TEXT
