Reproduce “DeepTx: Real-Time Transaction Risk Analysis via Multi-Modal Features and LLM Reasoning” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yixuan Liu, Xinlei Li, Yi Li
DeepTx is a system developed to analyze Web3 transaction risk in real-time, prior to user confirmation, by combining multi-modal features including behavioral, contextual, and UI signals with LLM reasoning and a consensus mechanism. In evaluations, it attained a precision of 1.00 and a recall of 0.89 (F1 score 0.94) in detecting phishing transactions, demonstrating the critical role of multi-modal data in its effectiveness.
Key insights
- Proactive, pre-signing analysis is crucial to counter Web3 phishing that exploits the semantic gap and UI deception, which reactive systems cannot address.
- Integrating multi-modal features, particularly UI signals and external threat intelligence, significantly enhances the detection capabilities against sophisticated attacks.
- Utilizing a multi-LLM consensus and self-reflection mechanism improves the robustness and explainability of risk assessments, mitigating LLM variability.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.