레이블 교체 방식 기반 지식그래프 부정 샘플링 및 지식 그래프 학습 모델 = Negative Sampling based on Label Swap for a Knowledge Graph

다국어 초록 (Multilingual Abstract)

As modern network infrastructures become increasingly complex, the need for advanced technologies and countermeasures is growing. Traditional fault analysis methods often rely on specific thresholds to predict anomalies, which do not fully leverage the structural information within the network. To address this limitation, we propose a method that converts time-series data into a knowledge graph to learn network relationship information. Negative sampling is employed to enhance the model's generalization performance by updating the features of each node and learning their interrelationships. This approach reduces the bias toward the 'normal' state, which is more prevalent in the data, and improves performance by exposing the model to a wider range of scenarios. Consequently, the ability to distinguish similar network faults has significantly improved, allowing for the consideration of various factors beyond immediate causes, leading to overall operational enhancements.