Transaction Anomaly Detection via Graph Embeddings
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional transaction processing systems face challenges in scaling anomaly detection for large transactions with many attributes and fail to identify interactions between related transactions, leading to inefficiencies and inaccuracies in detecting fraudulent behavior.
Innovation Solution
The use of graph embedding techniques to preprocess transaction data into a lower-dimensional space, combined with machine learning models, to detect anomalous behavior by capturing patterns in transaction networks and neighborhoods, employing TopkPPR and neighborhood-based embedding methods.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional anomaly detection methods are used to process transactions with large numbers of attributes, then detection accuracy may be maintained, but the system cannot scale efficiently and processing time increases significantly
Solution Approach 1:
The patent segments the high-dimensional transaction attribute space into multiple lower-dimensional subspaces using dimensionality reduction techniques. Each subspace captures specific patterns or features, allowing parallel processing and faster anomaly detection while maintaining comprehensive coverage of all transaction attributes.
Solution Approach 2:
The patent transforms the original high-dimensional transaction data into a lower-dimensional embedding space while preserving the essential relationships and patterns. This dimensionality change enables efficient processing and scaling without losing the ability to detect anomalies, as the embedded representation maintains the structural information needed for accurate detection.
2Measurement precision
If traditional transaction evaluation techniques process transactions in isolation, then individual transaction attributes can be evaluated, but interactions between related transactions are missed leading to inaccurate fraud detection
Solution Approach 1:
The patent merges information from multiple related transactions by constructing graphs that connect transactions through shared entities (users, merchants, devices). This combining approach allows the system to evaluate transactions in the context of their relationships, capturing fraud patterns that span multiple transactions while managing complexity through graph-based representation.
Solution Approach 2:
The patent introduces graph embeddings as an intermediary representation that captures the complex relationships between transactions. These embeddings serve as a mediator that encodes interaction patterns between related transactions, allowing the machine learning model to efficiently process relationship information without directly handling the full complexity of transaction graphs.
3Measurement precision
If more transaction attributes are analyzed to improve fraud detection, then detection accuracy improves, but the computational complexity and processing time increase
Solution Approach 1:
The patent applies dimensionality reduction to transform the high-dimensional attribute space into a lower-dimensional embedding space. This transformation preserves the essential information needed for accurate anomaly detection while reducing the computational complexity of processing. The embedded representation maintains the relationships between attributes without requiring direct processing of all original dimensions.
Solution Approach 2:
The patent changes the parameter representation from raw transaction attributes to graph embedding vectors. This parameter transformation allows the system to work with a fixed-dimensional representation regardless of the number of original attributes, enabling consistent processing complexity while capturing information from all input attributes through the embedding process.
Data Source
AI summary
Techniques are disclosed in which a computer system generates a transaction network graph from an initial set of transactions including known labels and attributes. The computer system may generate first and second matrices using first and second graph embedding routines from a training set of transactions that includes a first subset of transactions in the network graph. The first routine is based on anomalies in related transactions occurring at nodes in the transaction network graph that are multiple hops away while the second routine is based on anomalies in neighborhoods of similar transactions. In some embodiments, the computer system generates a final embedded matrix from the first and second matrices and uses the final matrix and a testing set of transactions that includes a second subset of transactions in the graph to train a machine learning model, where the trained model usable to determine whether unlabeled transactions are anomalous.