Graph Embedding for Similar Case Detection in Anomaly Investigation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current anomaly detection systems rely heavily on human investigators to analyze transactional data, which is tedious, time-consuming, and prone to errors, requiring significant expertise and resources.
Innovation Solution
Implement graph machine learning techniques, including graph embedding, shuffling, and unsupervised training, to automatically identify similar past case graphs and generate vector representations of vertices and edges, using improved GraphSAGE algorithms for enhanced accuracy and efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If human investigators manually analyze transactional data using predefined rules, then detection accuracy can be maintained through expert knowledge, but the investigation process becomes tedious, time-consuming, and resource-intensive
Solution Approach 1:
The system enables self-service anomaly detection through unsupervised graph machine learning models that automatically analyze transactional data without requiring human investigators to manually examine each case. The model independently generates anomaly scores and identifies suspicious patterns, freeing investigators from tedious manual analysis while maintaining detection accuracy through learned representations of normal and abnormal behavior
Solution Approach 2:
The patent replaces the mechanical system of manual human investigation with an automated graph machine learning system. Instead of investigators manually applying predefined rules and observing patterns, the system uses graph neural networks to automatically process transactional data, generate graph embeddings, and identify anomalies through learned representations, significantly reducing investigation time while maintaining reliability
2Reliability
If human investigators manually inspect cases, then expert knowledge can be applied to identify problematic behavior, but significant training resources and expertise maintenance are required
Solution Approach 1:
The graph machine learning model performs self-service learning by automatically training on historical transactional data to learn representations of normal and abnormal behavior patterns. The unsupervised learning approach enables the model to independently identify problematic behaviors without requiring human experts to manually label training data or continuously update their knowledge, eliminating the need for extensive investigator training while maintaining detection accuracy
3Productivity
If graph machine learning techniques are implemented to automate investigation, then investigation speed and accuracy improve, but computational complexity and data processing requirements increase
Solution Approach 1:
The patent segments the complex graph machine learning process into distinct computational stages: graph construction from transactional data, graph embedding generation using neural networks, graph shuffling for data augmentation, and anomaly scoring through comparison with training graphs. This segmentation allows each component to be optimized independently and processed efficiently, managing computational complexity while maintaining high investigation speed through parallel processing capabilities
Solution Approach 2:
The system performs preliminary actions by pre-training graph machine learning models on historical transactional data to learn representations of normal and abnormal patterns before actual anomaly detection. Graph embeddings are pre-computed and stored, allowing rapid inference on new transactions without requiring complex real-time computation, thus improving investigation speed while managing computational complexity through offline preparation
4Measurement precision
If vector representations of graphs are generated for similarity comparison, then case similarity identification improves, but storage requirements and computational overhead increase
Solution Approach 1:
The patent transforms complex graph structures into compact vector representations (graph embeddings) that capture essential structural and semantic information in a condensed format. By changing the representation parameters from detailed graph data to fixed-dimensional vectors, the system achieves precise similarity detection through vector comparison while significantly reducing storage requirements compared to storing complete graph structures for comparison
Data Source
AI summary
Herein is machine learning for anomalous graph detection based on graph embedding, shuffling, comparison, and unsupervised training techniques that can characterize an unfamiliar graph. In an embodiment, a computer obtains many known vectors that respectively represent known graphs. A new vector is generated that represents a new graph that contains multiple vertices. The new vector may contain an arithmetic aggregation of vertex vectors that respectively represent multiple vertices and/or a vector that represents a virtual vertex that is connected to the multiple vertices by respective virtual edges. In the many known vectors, some similar vectors that are similar to the new vector are identified. The new graph is automatically characterized based on a subset of the known graphs that the similar vectors represent.