Abstract
This study addresses the problem of network data analysis in 5G digital twin environments. To this end, network log data generated through simulations of real-world network failures were utilized to design a sophisticated log analysis module. The core techniques include graph embedding to capture the spatial structure of network data and the Fourier Transform algorithm to summarize and analyze log information. Subsequently, various hyperparameters, such as bandwidth settings, were adjusted to minimize noise across the entire log dataset and to extract meaningful log label types. Through this approach, the proposed method achieved a performance improvement of nearly 60% compared to conventional LLM-based embeddings, while also yielding clear and interpretable clustering results.