Scholarlink

The clustering of cybersecurity policy documents presents a significant challenge in legal Natural Language Processing (NLP), particularly within government and defense sectors. This study evaluates the effectiveness of clustering techniques when applied to cybersecurity policies represented using BERT-based embeddings. We employ dimensionality reduction methods, including Principal Component Analysis (PCA), t-Distributed Stochastic Neighbor Embedding (t-SNE), and Uniform Manifold Approximation and Projection (UMAP), to project high-dimensional embeddings into lower-dimensional space. We then assess the performance of K-Means, DBSCAN, and Hierarchical Clustering in organizing policy documents. Our results indicate that UMAP combined with Hierarchical Clustering achieves the highest clustering performance, attaining an Adjusted Rand Index (ARI) of 0.8529. These findings demonstrate the impact of transformer-based language models on cybersecurity policy analysis and highlight the role of dimensionality reduction in improving clustering effectiveness.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Cybersecurity Policy Clustering with LLM-Based Embeddings and Dimensionality Reduction

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Adriana García Aguirre,
  • Pablo Rivas,
  • Liang Sun

摘要

The clustering of cybersecurity policy documents presents a significant challenge in legal Natural Language Processing (NLP), particularly within government and defense sectors. This study evaluates the effectiveness of clustering techniques when applied to cybersecurity policies represented using BERT-based embeddings. We employ dimensionality reduction methods, including Principal Component Analysis (PCA), t-Distributed Stochastic Neighbor Embedding (t-SNE), and Uniform Manifold Approximation and Projection (UMAP), to project high-dimensional embeddings into lower-dimensional space. We then assess the performance of K-Means, DBSCAN, and Hierarchical Clustering in organizing policy documents. Our results indicate that UMAP combined with Hierarchical Clustering achieves the highest clustering performance, attaining an Adjusted Rand Index (ARI) of 0.8529. These findings demonstrate the impact of transformer-based language models on cybersecurity policy analysis and highlight the role of dimensionality reduction in improving clustering effectiveness.