Research
My research is in natural language processing and machine learning, with a focus on fake news detection in Turkish, text classification and comparing deep learning models with classical methods.
Full list and citations: Google Scholar · ORCID · ResearchGate · DergiPark · Hugging Face
Publications
- Fake News Detection on Mainstream Media Using Natural Language Processing
Abstract
The study proposes a linguistics-based model for detecting fake news in Turkish. On a dataset of news titles and summaries from 2015–2023, collected from the TRT RSS service and the Teyit.org fact-checking platform, six classical classifiers (SVM, logistic regression, random forest, k-NN, decision tree, Naive Bayes) are compared with an LSTM network, together with TF-IDF versus CountVectorizer representations and the effect of hyperparameter tuning. The best result, 93.12% accuracy, was obtained with an SVM using TF-IDF and tuning.
BibTeX
@article{kulaksiz2025fakenews, title = {Fake News Detection on Mainstream Media Using Natural Language Processing}, author = {Kulaksız, İsa and Coşkunçay, Ahmet}, journal = {Black Sea Journal of Engineering and Science}, year = {2025}, volume = {8}, number = {1}, pages = {214--224}, doi = {10.34248/bsengineering.1527551} } - Doğal Dil İşleme Kullanılarak Ana Akım Medyada Türkçe Sahte Haber Tespiti
Abstract
The thesis develops a system that classifies news from Turkish mainstream media as fake or real. News collected from TRT Gündem and Teyit.org is preprocessed and labelled; features are extracted with TF-IDF, CountVectorizer and Word2Vec; SVM, logistic regression, Naive Bayes, k-NN, decision tree, random forest, LSTM and BERTurk models are compared. The dataset is shared on Hugging Face as TR-FakeNews.
BibTeX
@mastersthesis{kulaksiz2025thesis, title = {Doğal Dil İşleme Kullanılarak Ana Akım Medyada Türkçe Sahte Haber Tespiti}, author = {Kulaksız, İsa}, school = {Atatürk Üniversitesi, Fen Bilimleri Enstitüsü, Bilgisayar Mühendisliği Anabilim Dalı}, year = {2025}, address = {Erzurum, Türkiye}, type = {Yüksek Lisans Tezi}, url = {https://tez.yok.gov.tr/UlusalTezMerkezi/} }
Open resources
Models and datasets I share on Hugging Face.
- TR-FakeNews
Turkish fake news dataset built from mainstream media: 5,325 news titles and summaries from 2015–2023 (sourced from TRT and Teyit.org), labelled real / fake. The data behind the journal article.
- turkish-pii-detection
Token classification model fine-tuned from BERTurk to detect and mask personal data (names, locations, organizations) in Turkish text, for KVKK / GDPR compliance and text anonymization.