FusionVul: A multimodal feature fusion framework for source code vulnerability detection

doi:10.1016/j.jss.2026.112992

FusionVul: A multimodal feature fusion framework for source code vulnerability detection

2026-06-08

0

PRE

AI

DOI：10.1016/j.jss.2026.112992

Original

Original request for help

Save

Abstract

En 中文

• Develop a hybrid framework combining sequential syntactic and structural semantics representations via UniXcoder and GGNN. • Design a Cross-Attention Feature Fusion Network (CAFFNet) for explicit cross-modal interaction and fine-grained feature alignment. • Propose a Sample-Aware Weighting (SaW) strategy for multi-semantic integration at the prediction stage.

FusionVul: A multimodal feature fusion framework for source code vulnerability detection

Abstract

Journal

Organization