부산대학교 도서관

Currently, Deep learning techniques are being investigated by researchers as a way to automatically detect smart contract flaws. This strategy seeks to get beyond the drawbacks of employing expert-defined patterns for detecting vulnerabilities in smart contracts, such as low detection rates and inefficiencies. However, The majority of recent research focuses on extracting features from smart contract code using a single code representation, such as an abstract syntax tree, control flow graph, or program dependency graph. These single code representations may lead to erroneous vulnerability detection and missing semantic information. This paper introduces a method called FBB-VD that uses a graph neural network to combine multiple code representations and detect vulnerabilities in smart contracts. This method can cover a wider range of code and detect vulnerabilities more accurately through more detailed features. The FBB-VD method is more successful and accurate in identifying vulnerabilities in smart contracts when compared to approaches that just transform smart contracts into abstract syntax trees, control flow graphs, or program dependency graphs.