Semantic Detection of Turkish Phishing Emails: A Natural Language Processing and Deep Learning-Based Approach

Merve Gül Taş

Makale detayları

Semantic Detection of Turkish Phishing Emails: A Natural Language Processing and Deep Learning-Based Approach

Yazarlar:

Merve Gül Taş

Yayıncı:

Duzce University

Ana Başlıklar:

Veri ve Bilgi Gizliliği | Doğal Dil İşleme

Dergi:

Siber Güvenlik ve Dijital Ekonomi Vol.1 Issue 1, pp.29-42

Belge Tipi:

Makale

Yayın Tarihi:

6/2025

Belge Dili:

Türkçe

Document ID (Org No.):

2001353913

pISSN:

3108-3706

Anahtar kelimeler:

Özet:

 This study proposes a semantic-based text classification approach for detecting phishing attacks in Turkish email content. A balanced dataset consisting of legitimate and fraudulent emails was constructed. The preprocessing phase included case normalization, punctuation removal, and TF-IDF-based vectorization, while contextual embeddings were obtained using the BERTurk model. Classification was performed using Naive Bayes, SVM, LSTM, ELM, and BERT algorithms. All models were trained in the Google Colab environment, and their performance was assessed using accuracy, F1-score, and ROC-AUC metrics. Results indicate that the BERT model provides superior performance in identifying semantic differences in Turkish phishing emails. The study addresses the gap in phishing detection for morphologically rich languages such as Turkish and presents a scalable model suitable for integration into real-time cybersecurity systems. The findings also demonstrate the viability of contextual NLP techniques in resource-scarce language environments.

 This study proposes a semantic-based text classification approach for detecting phishing attacks in Turkish email content. A balanced dataset consisting of legitimate and fraudulent emails was constructed. The preprocessing phase included case normalization, punctuation removal, and TF-IDF-based vectorization, while contextual embeddings were obtained using the BERTurk model. Classification was performed using Naive Bayes, SVM, LSTM, ELM, and BERT algorithms. All models were trained in the Google Colab environment, and their performance was assessed using accuracy, F1-score, and ROC-AUC metrics. Results indicate that the BERT model provides superior performance in identifying semantic differences in Turkish phishing emails. The study addresses the gap in phishing detection for morphologically rich languages such as Turkish and presents a scalable model suitable for integration into real-time cybersecurity systems. The findings also demonstrate the viability of contextual NLP techniques in resource-scarce language environments.

 This study proposes a semantic-based text classification approach for detecting phishing attacks in Turkish email content. A balanced dataset consisting of legitimate and fraudulent emails was constructed. The preprocessing phase included case normalization, punctuation removal, and TF-IDF-based vectorization, while contextual embeddings were obtained using the BERTurk model. Classification was performed using Naive Bayes, SVM, LSTM, ELM, and BERT algorithms. All models were trained in the Google Colab environment, and their performance was assessed using accuracy, F1-score, and ROC-AUC metrics. Results indicate that the BERT model provides superior performance in identifying semantic differences in Turkish phishing emails. The study addresses the gap in phishing detection for morphologically rich languages such as Turkish and presents a scalable model suitable for integration into real-time cybersecurity systems. The findings also demonstrate the viability of contextual NLP techniques in resource-scarce language environments.

 Bu çalışma, Türkçe e-posta içeriklerindeki kimlik avı phishing saldırılarını anlamsal düzeyde tespit etmeye yönelik bir metin sınıflandırma yaklaşımı sunmaktadır. Gerçek ve sahte e-postalardan dengeli olarak oluşturulmuş bir veri kümesi kullanılmıştır. Ön işleme sürecinde küçük harfe dönüştürme, noktalama işaretlerinin temizlenmesi ve TF-IDF tabanlı vektörleştirme uygulanmış; bağlamsal temsiller ise BERTurk modeli aracılığıyla elde edilmiştir. Sınıflandırma işlemi Naive Bayes, SVM, LSTM, ELM ve BERT algoritmalarıyla gerçekleştirilmiştir. Modeller Google Colab ortamında eğitilmiş ve doğruluk, F1 skoru ile ROC-AUC metrikleri üzerinden değerlendirilmiştir. Sonuçlar, BERT modelinin Türkçe phishing e-postalarındaki anlamsal farkları başarılı biçimde ayırt ettiğini ortaya koymaktadır. Bu çalışma, morfolojik açıdan zengin dillerde phishing tespiti konusunda literatürdeki boşluğu doldurmayı amaçlamakta ve gerçek zamanlı siber güvenlik sistemlerine entegre edilebilecek ölçeklenebilir bir model önermektedir. Elde edilen bulgular, düşük kaynaklı dillerde bağlamsal doğal dil işleme yöntemlerinin etkinliğini de ortaya koymaktadır.

Makale detayları

Semantic Detection of Turkish Phishing Emails: A Natural Language Processing and Deep Learning-Based Approach

Yazarlar:

Merve Gül Taş

Yayıncı:

Duzce University

Ana Başlıklar:

Veri ve Bilgi Gizliliği | Doğal Dil İşleme

Dergi:

Siber Güvenlik ve Dijital Ekonomi Vol.1 Issue 1, pp.29-42

Belge Tipi:

Makale

Yayın Tarihi:

6/2025

Belge Dili:

Türkçe

Document ID (Org No.):

2001353913

pISSN:

3108-3706

Deep Learning

Machine Learning

Natural Language Processing

Phishing

Turkish EmailsDerin Öğrenme

Doğal Dil İşleme

Kimlik Avı

Makine Öğrenmesi

Türkçe E-posta

Sayı

Makale detayları

Yazarlar: Merve Gül Taş

Yayıncı: Duzce University

Ana Başlıklar: Veri ve Bilgi Gizliliği | Doğal Dil İşleme

Dergi: Siber Güvenlik ve Dijital Ekonomi Vol.1 Issue 1, pp.29-42

Belge Tipi: Makale

Yayın Tarihi: 6/2025

Belge Dili: Türkçe

Document ID (Org No.): 2001353913

pISSN: 3108-3706

Sayı

Yazarlar:

Merve Gül Taş

Yayıncı:

Duzce University

Ana Başlıklar:

Veri ve Bilgi Gizliliği | Doğal Dil İşleme

Dergi:

Siber Güvenlik ve Dijital Ekonomi Vol.1 Issue 1, pp.29-42

Belge Tipi:

Makale

Yayın Tarihi:

6/2025

Belge Dili:

Türkçe

Document ID (Org No.):

2001353913

pISSN:

3108-3706