Topic Clustering

Topic data not available yet.

Topic clustering data will appear once the BERTopic pipeline runs.

Methodology

Documents are embedded using Legal-BERT(nlpaueb/legal-bert-base-uncased), a BERT model fine-tuned on legal text. Clustering is performed using HDBSCAN with UMAP dimensionality reduction. Topic terms are extracted usingc-TF-IDF (class-based Term Frequency-Inverse Document Frequency).

  • Embedding model: Legal-BERT (768 dimensions)
  • Clustering: HDBSCAN (min_cluster_size=3, min_samples=2)
  • Dimensionality reduction: UMAP (n_neighbors=15, n_components=5)
  • Topic extraction: c-TF-IDF with bigram support