Influential Node Identification With QUBO Graph Clustering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing graph influence models fail to account for unique object behaviors in real-world networks, such as social networks, leading to inaccurate predictions of influence propagation and neglecting factors like social pressure or user interface effects.
Innovation Solution
Model influence maximization as a discrete optimization problem using a quadratic unconstrained binary optimization (QUBO) function to identify influential nodes by partitioning graph datasets into balanced clusters and applying quantum annealing techniques to select top influencer nodes based on cumulative influence.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional graph influence models are used to identify influential nodes, then the model is simple to implement, but the accuracy of influence propagation prediction is poor because unique object behaviors in real-world networks are not accounted for
Solution Approach 1:
The patent segments the graph dataset into multiple balanced clusters using spectral clustering, then applies quantum annealing within each cluster to identify influential nodes. This segmentation approach allows the model to account for local network-specific behaviors while maintaining computational tractability, thereby improving prediction accuracy without excessive complexity
Solution Approach 2:
The patent transforms the influence maximization problem into a quadratic unconstrained binary optimization (QUBO) formulation, changing the parameter representation from traditional continuous influence scores to binary selection variables. This parameter transformation enables the use of quantum annealing and digital annealing techniques that can capture complex network behaviors while maintaining a structured optimization framework
2Measurement precision
If quantum annealing techniques are applied to identify influential nodes, then the accuracy of identifying influential nodes is improved, but the computational complexity increases
Solution Approach 1:
The patent divides the large graph dataset into multiple balanced clusters before applying quantum annealing to each cluster separately. This segmentation reduces the computational burden on quantum annealing solvers by working with smaller subproblems, thereby improving the efficiency-time tradeoff while maintaining identification accuracy
Solution Approach 2:
The patent applies quantum annealing selectively to identify top-k influential nodes within each cluster rather than computing influence scores for all nodes. This partial action approach focuses computational resources on the most critical nodes, achieving high accuracy for the most influential nodes without the excessive computational cost of full network analysis
3Measurement precision
If graph datasets are partitioned into balanced clusters, then the identification of influential nodes within clusters is improved, but the overall processing time increases due to multiple processing steps
Solution Approach 1:
The patent uses spectral clustering to partition the graph into balanced clusters that preserve network structure while enabling independent processing. This segmentation improves identification accuracy by capturing local community structures, and the parallel processing capability of multiple clusters helps offset the additional processing time through concurrent computation
Solution Approach 2:
The patent performs preliminary spectral clustering to create balanced clusters before applying the computationally intensive quantum annealing process. This preliminary action organizes the data in a way that facilitates more efficient subsequent processing, reducing the overall processing time by avoiding repeated clustering operations during the influence identification phase
Data Source
AI summary
A method may include obtaining a graph dataset that represents a social network and includes a plurality of nodes. The method may include partitioning the graph dataset into a first node cluster and a second node cluster in which the first node cluster and the second node cluster each include one or more nodes from the graph dataset. The method may include identifying one or more first candidate influential nodes included in the first node cluster and one or more second candidate influential nodes included in the second node cluster. The method may include selecting one or more of the first candidate influential nodes and the second candidate influential nodes as influencer nodes and identifying a respective object included in the social network corresponding to each of the influencer nodes.