Graph Representation Learning via Partial Graph Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies face challenges in efficiently calculating representation vectors for large datasets due to resource constraints, particularly when dealing with vast numbers of users and products, leading to incomplete capture of relationships and decreased performance.
Innovation Solution
The approach involves dividing an entire graph representing relationships into smaller partial graphs, setting objective functions for each, and calculating representation vectors for these partial graphs, which are then combined to form a final vector, incorporating inter-graph relationships and utilizing knowledge graphs to enhance accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If representation vectors are calculated for all users and products in the entire graph, then the completeness of relationship capture is improved, but the calculation resource requirements become excessive
Solution Approach 1:
The patent divides the entire graph into multiple partial graphs based on connectivity thresholds. Representation vectors are calculated separately for each partial graph, which contains fewer nodes and edges. This segmentation reduces the computational complexity from O(V_total × E_total) to sum of O(V_i × E_i) for each partial graph i, making the calculation feasible while preserving important local relationships.
2Adaptability or versatility
If the number of users and products increases, then the coverage of relational data is improved, but the size of representation vector becomes enormous
Solution Approach 1:
By segmenting the graph into partial graphs with fewer nodes, the representation vector size for each partial graph is significantly reduced. The patent calculates representation vectors only for nodes within each partial graph, avoiding the need to compute enormous vectors for the entire graph while maintaining adaptability to large datasets through systematic division.
3Reliability
If representation vectors are calculated for large datasets, then the comprehensiveness of data analysis is improved, but calculation performance decreases
Solution Approach 1:
The patent improves calculation performance by dividing the large dataset into smaller partial graphs that can be processed in parallel. Each partial graph is independently processed, reducing the computational burden per unit and enabling faster execution while maintaining comprehensive analysis through the aggregation of results from all partial graphs.
Data Source
AI summary
An information processing device includes at least one hardware processor. The hardware processor is configured to divide an entire graph indicating a relationship between a plurality of elements into a plurality of partial graphs. The hardware processor is configured to set an objective function for learning a representation vector of a partial graph for each of the plurality of partial graphs. The hardware processor is configured to calculate n representation vectors by learning to optimize the objective function using each of n (n is an integer of 1 or more) setting values for each of the plurality of partial graphs.