Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Dynamic Feature Fusion: Combining Global Graph Structures and Local Semantics for Blockchain Fraud Detection

Zhang Sheng, Liangliang Song, Yanbin Wang

TL;DR

The paper tackles blockchain fraud detection by jointly modeling global graph structures and local semantic information from transaction data. It introduces ETH-GBERT, a dynamic multimodal fusion model that merges a GCN-based graph representation with a BERT-based semantic encoder through a gating mechanism, enabling adaptive fusion across inputs. A comprehensive data processing pipeline constructs the transaction graph, enriches temporal features with n-gram time differences, and preprocesses text data for semantic modeling, leading to state-of-the-art results on large Ethereum datasets. The work demonstrates that adaptive, modality-aware fusion of structural and semantic cues significantly improves detection accuracy, robustness, and scalability for securing blockchain systems.

Abstract

The advent of blockchain technology has facilitated the widespread adoption of smart contracts in the financial sector. However, current fraud detection methodologies exhibit limitations in capturing both global structural patterns within transaction networks and local semantic relationships embedded in transaction data. Most existing models focus on either structural information or semantic features individually, leading to suboptimal performance in detecting complex fraud patterns.In this paper, we propose a dynamic feature fusion model that combines graph-based representation learning and semantic feature extraction for blockchain fraud detection. Specifically, we construct global graph representations to model account relationships and extract local contextual features from transaction data. A dynamic multimodal fusion mechanism is introduced to adaptively integrate these features, enabling the model to capture both structural and semantic fraud patterns effectively. We further develop a comprehensive data processing pipeline, including graph construction, temporal feature enhancement, and text preprocessing. Experimental results on large-scale real-world blockchain datasets demonstrate that our method outperforms existing benchmarks across accuracy, F1 score, and recall metrics. This work highlights the importance of integrating structural relationships and semantic similarities for robust fraud detection and offers a scalable solution for securing blockchain systems.

Dynamic Feature Fusion: Combining Global Graph Structures and Local Semantics for Blockchain Fraud Detection

TL;DR

The paper tackles blockchain fraud detection by jointly modeling global graph structures and local semantic information from transaction data. It introduces ETH-GBERT, a dynamic multimodal fusion model that merges a GCN-based graph representation with a BERT-based semantic encoder through a gating mechanism, enabling adaptive fusion across inputs. A comprehensive data processing pipeline constructs the transaction graph, enriches temporal features with n-gram time differences, and preprocesses text data for semantic modeling, leading to state-of-the-art results on large Ethereum datasets. The work demonstrates that adaptive, modality-aware fusion of structural and semantic cues significantly improves detection accuracy, robustness, and scalability for securing blockchain systems.

Abstract

The advent of blockchain technology has facilitated the widespread adoption of smart contracts in the financial sector. However, current fraud detection methodologies exhibit limitations in capturing both global structural patterns within transaction networks and local semantic relationships embedded in transaction data. Most existing models focus on either structural information or semantic features individually, leading to suboptimal performance in detecting complex fraud patterns.In this paper, we propose a dynamic feature fusion model that combines graph-based representation learning and semantic feature extraction for blockchain fraud detection. Specifically, we construct global graph representations to model account relationships and extract local contextual features from transaction data. A dynamic multimodal fusion mechanism is introduced to adaptively integrate these features, enabling the model to capture both structural and semantic fraud patterns effectively. We further develop a comprehensive data processing pipeline, including graph construction, temporal feature enhancement, and text preprocessing. Experimental results on large-scale real-world blockchain datasets demonstrate that our method outperforms existing benchmarks across accuracy, F1 score, and recall metrics. This work highlights the importance of integrating structural relationships and semantic similarities for robust fraud detection and offers a scalable solution for securing blockchain systems.
Paper Structure (34 sections, 10 equations, 4 figures, 3 tables)

This paper contains 34 sections, 10 equations, 4 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Graph-based Fraud Detection
  4. Fraud Detection Based on Time Series Data
  5. Hybrid Methods
  6. Methodology
  7. Data generation and pre-processing
  8. Time Aggregation Feature Enhancement
  9. Graph Data Generation
  10. Text Transaction Data Generation
  11. Text Data Cleaning
  12. ETH-GBERT Model Architecture
  13. Model Architecture
  14. Graph-Based Representation Module Design
  15. Semantic Feature Extraction Module Design
  16. ...and 19 more sections

Figures (4)

  • Figure 1: Architecture of the Dynamic Feature Fusion Model for Blockchain Fraud Detection
  • Figure 2: 3D Comparative Analysis of Model Performance Metrics Across Datasets
  • Figure 3: Heatmap of Model Performance Metrics Across Multiple Datasets
  • Figure 4: Trends of F1 Score, Recall, and Precision Across Different Normal-to-Fraud Ratios