erdem-erdem/Turkish-Law-Documents-700k-clustered

Turkish Legal Documents Clustering Dataset

A comprehensive dataset of 700,000 Turkish legal documents from the two primary sources of legal precedent in Turkey, clustered using multiple emebdding models and algorithms to enable research, analysis, and machine learning applications.

Overview

This repository contains a large-scale document clustering pipeline and dataset for Turkish legal documents sourced from:

• —Yargıtay - Turkey's highest court of appeal for civil and criminal cases
• —Danıştay - Turkey's highest administrative court

These institutions are the authoritative sources of legal precedent and jurisprudence in the Turkish legal system. The documents represent decisions, verdicts, and judicial rulings spanning multiple years.

Dataset Features

Each record in the dataset contains:

• —text: The full legal document text in Turkish
• —source: Source identifier (ygty1, ygty2, dnsy1, dnsy2). These correspond to following datasets, respectively: fikriokan/ygty, fikriokan/ygty-2, fikriokan/dnsy-1, fikriokan/dnsy-2.
• —tr_e5_knn_cluster_id: Cluster ID from KNN clustering on Turkish E5 embeddings
• —orig_e5_knn_cluster_id: Cluster ID from KNN clustering on Multilingual E5 embeddings
• —tr_e5_hdbscan_cluster_id: Cluster ID from HDBSCAN clustering on Turkish E5 embeddings
• —orig_e5_hdbscan_cluster_id: Cluster ID from HDBSCAN clustering on Multilingual E5 embeddings

Data Source Attribution

• —Yargıtay: https://karararama.yargitay.gov.tr/
• —Danıştay: https://kararara.danistay.gov.tr/

The original legal documents are public records from Turkish government institutions and are provided in accordance with Turkish laws on freedom of information.