umutertugrul/turkish-academic-theses-dataset

📚 Turkish Academic Theses Abstracts (TR/EN)

A large-scale bilingual (Turkish–English) collection of abstracts from Turkish academic theses (YÖK Ulusal Tez Merkezi). This dataset focuses only on abstracts, provided in Turkish (abstract_tr) and English (abstract_en), suitable for summarization, translation, classification, and retrieval.

• Records: ~650k abstracts (TR & EN)
• Format: Parquet (.parquet) — fast & analytics-friendly
• Language: 🇹🇷 Turkish + 🇬🇧 English (parallel abstracts when available)

Note: This release contains abstracts only. Full texts are not included.

---

📐 Schema

Each row follows this structure (example keys and meaning):

Column	Type	Description
tez_no	integer	Thesis identifier (numeric code on YÖK; stable key)
pdf_url	string	Public handle page for the thesis PDF on YÖK (may require permissions/login for full text)
title_tr	string	Thesis title (Turkish)
title_en	string	Thesis title (English, if available)
author	string	Author’s full name
advisor	string	Advisor / supervisor (with title when present)
location	string	Institution path (university / institute / department / program)
subject	string	Subject area (often TR = EN format)
index	string	Index terms / keywords (often TR = EN ; …)
status	string	Thesis status (e.g., Onaylandı)
degree	string	Degree type (e.g., Yüksek Lisans, Doktora, ...)
language	string	Thesis language (Türkçe)
year	integer	Year of publication (e.g., 2016)
pages	integer	Page count (numeric string; sometimes missing)
abstract_tr	string	Turkish abstract (may be empty for a subset)
abstract_en	string	English abstract (may be empty for a subset)

Some entries may miss the English or Turkish abstract; always check for null/empty values.

---

📦 File Format

• Parquet (.parquet) → optimized for analytics and ML pipelines
• Fast to load with pandas or Hugging Face datasets

---

🔍 Use Cases

• Summarization (TR-only or EN-only)
• Translation (TR↔EN; parallel abstracts)
• Text Classification (e.g., research area labels after your own weak/strong labeling)
• Information Retrieval (semantic search over abstracts; bilingual IR)

---

🚀 Example Usage

Load the entire dataset

from datasets import load_dataset

ds = load_dataset("umutertugrul/turkish-academic-theses-dataset")
print(ds)

Load as a DataFrame (pandas)

import pandas as pd

df = pd.read_parquet("tezler.parquet")
print(df.head())

Load with filtering (e.g., theses from 2020)

from datasets import load_dataset

ds = load_dataset("umutertugrul/turkish-academic-theses-dataset", split="train")
df = ds.to_pandas()
df_2020 = df[df["year"] == 2020]
print(df_2020.head())

---

📥 Citation

If you use this dataset in your research or project, please cite:

title = Turkish Academic Theses Abstracts (TR/EN) author = Umut Ertuğrul Daşgın url = https://huggingface.co/umutertugrul