Community Search Model for Overlapping Graph Structures
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing community search methods struggle with overlapping community structures in complex networks, where a node can belong to multiple communities, leading to inefficiencies and inaccuracies in identifying community memberships.
Innovation Solution
A method and apparatus for community search using an electronic device that repeatedly determines query nodes and performs community searches within a graph data structure, allowing nodes to belong to multiple communities by adding closely connected nodes to the target community, utilizing a community search model that includes graph neural networks and reinforcement learning to optimize community identification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional community search methods are used, then the community search process is simple, but the accuracy of identifying overlapping community memberships deteriorates
Solution Approach 1:
The patent segments the community search process into two distinct phases: a training phase where the model learns community structures from labeled data, and an inference phase where the trained model performs community search. This segmentation allows the complex learning process to be separated from the search process, improving accuracy while managing complexity through model reuse.
Solution Approach 2:
The patent performs preliminary training of the community search model using labeled graph data before actual community search operations. The model learns community structures, node representations, and community assignments in advance, so that during inference, the search can be performed accurately without repeating the complex learning process.
2Measurement precision
If community search is performed for each node individually, then the community membership is accurately determined, but the computational workload increases
Solution Approach 1:
The patent creates a universal community search model that can perform community search for multiple nodes simultaneously. The trained model learns general community structures from the graph data and can apply this knowledge to determine communities for any query node, eliminating the need for individual node-by-node analysis while maintaining accuracy.
Solution Approach 2:
The patent uses graph neural networks to create vector representations (embeddings) of nodes and communities. Once the model is trained, it can efficiently compute community assignments for new nodes by comparing their vector representations to learned community prototypes, avoiding the need to reprocess the entire graph structure for each query node.
3Measurement precision
If the community search model is trained with comprehensive data, then the model accuracy improves, but the training time and resources increase
Solution Approach 1:
The patent uses a small subset of labeled graph data for training the community search model, rather than requiring comprehensive labeling of all nodes and communities. This partial action approach achieves sufficient model accuracy for practical applications while significantly reducing the time and resources needed for training compared to complete data labeling.
Data Source
AI summary
A method for community search is performed by an electronic device. The method includes: obtaining graph data to be processed, in which the graph data includes a plurality of nodes and a plurality of connection edges between the nodes; determining a query node from the plurality of nodes based on the graph data, and determining a target community to which the query node belongs by performing a community search for the query node, in which the target community includes the query node, and at least one node other than the query node in the plurality of nodes; and determining the query node and performing the community search for the query node repeatedly until the community to which each node included in the graph data belongs is determined.