Graph Neural Node Classification With Distribution-Based Sampling

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Graph neural networks face high calculation complexity and low prediction accuracy due to aggregating features of numerous neighboring nodes, making it difficult to meet application requirements.

Innovation Solution

A node classification method that involves obtaining category distribution of neighboring nodes through prediction, performing sampling based on this distribution to select sampled nodes, and using these nodes for inference, thereby reducing calculation complexity and improving accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If features of all neighboring nodes are aggregated, then prediction accuracy should be improved, but calculation complexity becomes very high

Engineering Contradiction:
Improveprediction accuracyVSAvoidcalculation complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent extracts only the necessary information from neighboring nodes by predicting category distribution and sampling representative nodes, rather than aggregating all neighboring node features. This extraction approach reduces calculation complexity while maintaining prediction accuracy by focusing on the most informative samples.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent creates a simplified representation of neighboring nodes through sampling. Instead of processing all original neighboring node features, it generates a subset of sampled nodes that replicate the essential category distribution characteristics, thereby reducing computational burden while preserving predictive power.

Inventive Principle:
Principle #26Copying

2Measurement precision

If features of all neighboring nodes are aggregated, then prediction accuracy should be improved, but the model becomes difficult to meet application requirements

Engineering Contradiction:
Improveprediction accuracyVSAvoidapplication feasibility
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The patent applies partial action by processing only a sampled subset of neighboring nodes rather than all neighboring nodes. This partial processing approach makes the model feasible for practical applications with large graphs, as it reduces computational requirements while still achieving acceptable prediction accuracy through representative sampling.

Inventive Principle:
Principle #16Partial or excessive action

3Device complexity

If sampling is performed based on labeled neighboring nodes, then calculation complexity is reduced, but prediction accuracy deteriorates due to sample imbalance

Engineering Contradiction:
Improvecalculation complexityVSAvoidprediction accuracy
Core Design Contradiction:
Device complexityVSMeasurement precision

Solution Approach 1:

The patent uses feedback from the node classification model to predict category distribution of unlabeled neighboring nodes. This feedback mechanism allows the system to iteratively refine the category distribution estimates, ensuring that sampled nodes accurately represent the true category distribution and avoiding sample imbalance issues.

Inventive Principle:
Principle #23Feedback

Solution Approach 2:

The patent changes the parameter used for sampling from simple label-based selection to probability-based sampling using predicted category distribution. This parameter change enables the system to sample nodes according to their true category probabilities, thereby maintaining prediction accuracy while reducing calculation complexity.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP4726579A1Node classification method and system based on graph neural network, and related device
Publication Date: 2026.04.15 HUAWEI TECH CO LTD
  • EP4726579A1 patent drawingFigure 1
  • EP4726579A1 patent drawingFigure 2
  • EP4726579A1 patent drawingFigure 3

AI summary

A graph neural network-based node classification method and system, and a related device are provided. In the method, during training of a model, categories of a plurality of neighboring node samples in a graph data sample are first predicted to obtain category distribution of the plurality of neighboring node samples, and sampling is then performed on the plurality of neighboring node samples based on the category distribution and a sampling parameter input by a user to obtain a plurality of sampled nodes, so that category distribution of the plurality of sampled nodes is similar to or consistent with the category distribution of the plurality of neighboring node samples. In this way, features of the sampled nodes obtained through sampling can cover features of all neighboring nodes, thereby reducing calculation complexity. In addition, the category distribution of the sampled nodes is closer to true distribution of the neighboring nodes, thereby improving performance of a graph neural network.