Transaction Data Graph Embedding Spatial-Temporal Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
High-dimensional and sparse transaction graphs in financial data analysis limit the utility for tasks like credit risk modeling and fraud detection, as existing graph embedding techniques fail to consider spatial-temporal information.
Innovation Solution
The method involves obtaining transaction data across multiple time windows, using a graph convolutional network and a transformer framework to extract spatial-temporal information, and generating feature representations for accounts based on this information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If transaction data is represented as a high-dimensional graph, then the graph can capture complex transaction relationships, but the graph becomes very sparse and limits utility for financial data analysis
Solution Approach 1:
The patent extracts spatial information from the high-dimensional graph structure using graph convolutional networks, separating it from temporal information. This extraction process transforms the sparse, high-dimensional graph data into denser spatial feature representations that can be effectively combined with temporal features, resolving the contradiction between capturing complex relationships and maintaining analytical utility
Solution Approach 2:
The patent transforms the spatial-temporal transaction data into a multi-dimensional feature space by combining spatial features (from graph structure) with temporal features (from time window sequences). This dimensionality transformation creates comprehensive feature representations that overcome the sparsity limitation while preserving the complex relationship information
2Ease of manufacture
If existing graph embedding techniques are used, then the graph structure can be processed, but spatial-temporal information is not considered
Solution Approach 1:
The patent merges graph convolutional networks (for spatial processing) with transformer models (for temporal processing) into a unified spatial-temporal processing framework. This combination allows the system to simultaneously process graph structure information and temporal sequence information, preventing loss of either spatial or temporal characteristics while maintaining ease of graph processing
Solution Approach 2:
The patent introduces dynamic temporal modeling by processing transaction data across multiple time windows and capturing temporal dependencies through the transformer framework. This dynamic approach adapts to changing transaction patterns over time, preserving temporal information that static graph embedding techniques would lose
Data Source
AI summary
A dynamic graph embedding method for transaction data analysis includes obtaining transaction data associated with an account during a plurality of time windows, extracting spatial-temporal information of the transaction data by using a graph convolutional network and a transformer framework, and generating a feature representation for the account based on the spatial-temporal information.