Graph Neural Network Node Feature Extraction via Structural Loss
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current graph analysis technologies face challenges in training networks for unsupervised learning due to the lack of training labels in friendship and co-selling relationship data, as contrastive loss methods fail to effectively utilize graph structure and properties.
Innovation Solution
An information processing apparatus that generates new graph data by edge erasure and node masking, calculates contrastive and structural losses using graph neural networks, and updates the extraction model to optimize node feature extraction, incorporating structural loss to better utilize graph information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If contrastive loss is used for graph self-supervised learning, then node feature extraction is enabled without training labels, but graph structure and property information is not fully utilized
Solution Approach 1:
The loss function is segmented into two distinct components: contrastive loss for node feature extraction and structural loss for graph structure preservation. This segmentation allows each loss component to focus on specific aspects, with structural loss explicitly capturing graph structure information that contrastive loss alone overlooks.
Solution Approach 2:
The patent merges contrastive loss and structural loss into a unified training objective. By combining these two loss functions, the system simultaneously achieves node feature extraction capability while fully utilizing graph structure and property information, resolving the information loss problem.
2Device complexity
If all negative examples are treated equally in contrastive loss, then computation is simplified, but graph structure and property information is not effectively utilized
Solution Approach 1:
The structural loss component introduces local quality differentiation by computing losses based on node degrees and graph distances. Different nodes and edges are treated with different weights according to their structural importance, allowing the system to capture graph structure information while maintaining manageable computational complexity.
Data Source
AI summary
According to one embodiment, an information processing apparatus includes a processor. The processor modifies graph data to generate two pieces of new graph data. The processor extracts, by an extraction model, features of respective nodes in the graph data. The processor calculates a contrastive loss by using similarity between the extracted features. The processor calculates a structural loss by using between the extracted features. The processor updates the extraction model using the contrastive loss and the structural loss.