Augmented Graph Modularity Maximization for Transaction Network Subsets
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Analyzing transactions between user accounts to identify subsets with shared characteristics is complex due to large-scale networks, heterogeneous data types, missing information, and contradictory topological and attribute information, making it challenging for existing methods to effectively detect meaningful communities for security, risk management, and marketing applications.
Innovation Solution
The Augmented Graph with Modularity Maximization and Refinement (AGGMMR) framework partitions an augmented graph model using both attribute and topological information through a greedy modularity maximization algorithm, consisting of three phases: augmented graph construction, weight learning, and modularity refinement, to determine subsets of user accounts.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional community detection methods are used on large-scale transaction networks, then computational complexity increases and processing time extends, but the ability to identify meaningful subsets of user accounts deteriorates
Solution Approach 1:
The patent segments the large-scale transaction network into multiple communities using the AGGMMR framework, which divides the graph into meaningful subsets based on modularity optimization. This segmentation approach allows efficient processing of large networks by breaking them into manageable communities while maintaining detection accuracy through the refined modularity function that accounts for node attributes and edge weights.
Solution Approach 2:
The patent changes the parameters of the modularity function by incorporating node attributes and edge weights into the objective function. The refined modularity Q includes attribute-based terms that weight the contribution of different node pairs based on their attribute similarity, allowing the algorithm to prioritize meaningful connections while maintaining computational efficiency on large-scale networks.
2Reliability
If both attribute information and topological information are used for community detection, then the completeness of community characteristics improves, but information conflicts and processing complexity increase
Solution Approach 1:
The patent merges attribute information and topological information into a unified modularity objective function. The refined modularity Q combines both types of information by incorporating attribute similarity weights into the edge weight calculations, allowing simultaneous consideration of network structure and node characteristics without requiring separate processing steps or complex multi-objective optimization.
Solution Approach 2:
The patent uses edge weights as an intermediary that mediates between topological connections and attribute similarity. The edge weights are computed as functions of both the topological relationship (whether nodes are connected) and the attribute similarity (how similar node attributes are), serving as a bridge that integrates both information types into the modularity optimization process in a computationally efficient manner.
3Adaptability or versatility
If greedy modularity maximization is applied to augmented graphs with heterogeneous nodes, then the handling of diverse data types improves, but computational resource utilization increases
Solution Approach 1:
The patent creates a universal framework that handles heterogeneous node types and diverse data types through a single augmented graph representation and unified modularity objective function. The framework accommodates different node attributes (categorical, numerical, textual) and edge types (transactions, interactions) within the same mathematical formulation, allowing the greedy optimization algorithm to process diverse data without requiring type-specific processing logic or multiple separate algorithms.
Data Source
AI summary
A computer system receives a first set of information that describes a set of transactions between pairs of user accounts of a service and a second set of information that specifies attribute values for user accounts of the service that are recipient user accounts within the set of transactions. Using the first set of information and the second set of information, the computer system generates an augmented graph model of the transactions and attribute values. By applying modularity maximization to the augmented graph model, the computer system determines a plurality of subsets of recipient user accounts.