Detecting Spams on the X Platform with Machine Learning, Large Language Models, and Computer Vision

Ebutalha Camadan; Mehmet Şimşek

Makale detayları

Detecting Spams on the X Platform with Machine Learning, Large Language Models, and Computer Vision

Yazarlar:

Ebutalha Camadan | Mehmet Şimşek

Yayıncı:

Gazi University

Ana Başlıklar:

Derin Öğrenme | Yarı ve Denetimsiz Öğrenme | Doğal Dil İşleme

Dergi:

Bilişim Teknolojileri Dergisi Vol.18 Issue 2, pp.181-194

Belge Tipi:

Makale

Yayın Tarihi:

4/2025

Belge Dili:

Türkçe

Document ID (Org No.):

2001345766

pISSN:

1307-9697

Anahtar kelimeler:

Özet:

 While social media platforms have become crucial tools for information sharing and communication, the spread of unwanted content spam has also become a significant problem. This paper proposes a novel approach for spam detection on the social media platform X formerly Twitter by integrating machine learning, large language models, and computer vision techniques. A dataset containing posts with visual content on popular Turkish topics was created, aiming to identify the most effective machine learning algorithms for spam detection. Feature engineering was conducted to capture key aspects of social media interaction, including the relationship between post content and hashtags, as well as the relevance between multiple hashtags. Additionally, image-based features were introduced, such as the initial posting date of an image on X and its textual similarity to other web pages, to enhance visual content analysis. These features were developed using Google Gemini and Cloud Vision AI. Experimental evaluations with five machine learning algorithms Decision Trees, Random Forest, SVM, Logistic Regression, and Multilayer Perceptron demonstrated that the Random Forest algorithm achieved the highest accuracy and F1 score. This paper highlights the effectiveness of machine learning methods in spam detection on X and introduces new methodologies for leveraging Google Gemini and Cloud Vision AI. Furthermore, the engineered features provide a strong foundation for accurately classifying spam content.

 While social media platforms have become crucial tools for information sharing and communication, the spread of unwanted content spam has also become a significant problem. This paper proposes a novel approach for spam detection on the social media platform X formerly Twitter by integrating machine learning, large language models, and computer vision techniques. A dataset containing posts with visual content on popular Turkish topics was created, aiming to identify the most effective machine learning algorithms for spam detection. Feature engineering was conducted to capture key aspects of social media interaction, including the relationship between post content and hashtags, as well as the relevance between multiple hashtags. Additionally, image-based features were introduced, such as the initial posting date of an image on X and its textual similarity to other web pages, to enhance visual content analysis. These features were developed using Google Gemini and Cloud Vision AI. Experimental evaluations with five machine learning algorithms Decision Trees, Random Forest, SVM, Logistic Regression, and Multilayer Perceptron demonstrated that the Random Forest algorithm achieved the highest accuracy and F1 score. This paper highlights the effectiveness of machine learning methods in spam detection on X and introduces new methodologies for leveraging Google Gemini and Cloud Vision AI. Furthermore, the engineered features provide a strong foundation for accurately classifying spam content.

 While social media platforms have become crucial tools for information sharing and communication, the spread of unwanted content spam has also become a significant problem. This paper proposes a novel approach for spam detection on the social media platform X formerly Twitter by integrating machine learning, large language models, and computer vision techniques. A dataset containing posts with visual content on popular Turkish topics was created, aiming to identify the most effective machine learning algorithms for spam detection. Feature engineering was conducted to capture key aspects of social media interaction, including the relationship between post content and hashtags, as well as the relevance between multiple hashtags. Additionally, image-based features were introduced, such as the initial posting date of an image on X and its textual similarity to other web pages, to enhance visual content analysis. These features were developed using Google Gemini and Cloud Vision AI. Experimental evaluations with five machine learning algorithms Decision Trees, Random Forest, SVM, Logistic Regression, and Multilayer Perceptron demonstrated that the Random Forest algorithm achieved the highest accuracy and F1 score. This paper highlights the effectiveness of machine learning methods in spam detection on X and introduces new methodologies for leveraging Google Gemini and Cloud Vision AI. Furthermore, the engineered features provide a strong foundation for accurately classifying spam content.

 Sosyal medya platformları, günümüzde bilgi paylaşımı ve iletişimde önemli araçlar haline gelirken, aynı zamanda istenmeyen gönderilerin spam yayılması da büyük bir sorun teşkil etmektedir. Bu çalışma, X sosyal medya platformundaki eski adıyla Twitter istenmeyen gönderilerin tespitine yönelik, makine öğrenmesi, geniş dil modelleri ve bilgisayarlı görü tekniklerini birleştiren yeni bir yaklaşım önermektedir. Türkiyede popüler olan konulara dair görsel içeren gönderilerden bir veri kümesi oluşturularak, spam tespitinde en etkili makine öğrenmesi algoritmaları belirlenmeye çalışılmıştır. Gönderi içeriğinin etiketlerle ilişkisi ve birden fazla etiketin birbiriyle ilgisi gibi sosyal medya etkileşimini belirleyen öznitelikler geliştirilmiştir. Ayrıca, görsel içeriğin analizi için, görselin X platformunda ilk paylaşıldığı tarih ile internet üzerindeki diğer sayfalarda geçtiği metinle benzerliği gibi görsel odaklı öznitelikler de dahil edilmiştir. Bu öznitelikler, Google Gemini ve Cloud Vision AI araçları kullanılarak geliştirilmiştir. Beş farklı makine öğrenmesi algoritması Karar Ağaçları, Rastgele Orman, SVM, Lojistik Regresyon, Çok Katmanlı Algılayıcı ile yapılan deneylerde, Rastgele Orman algoritması en yüksek doğruluk ve F1 skoru değerlerine ulaşmıştır. Bu çalışma, X platformunda istenmeyen gönderi tespiti için makine öğrenmesi yöntemlerinin etkinliğini göstermiş ve Google Gemini ile Cloud Vision AI araçlarının etkin kullanımına dair yeni yöntemler sunmuştur. Ayrıca geliştirilen öznitelikler, spam içeriklerin doğru bir şekilde sınıflandırılmasında güçlü bir temel oluşturmaktadır.

Makale detayları

Detecting Spams on the X Platform with Machine Learning, Large Language Models, and Computer Vision

Yazarlar:

Ebutalha Camadan | Mehmet Şimşek

Yayıncı:

Gazi University

Ana Başlıklar:

Derin Öğrenme | Yarı ve Denetimsiz Öğrenme | Doğal Dil İşleme

Dergi:

Bilişim Teknolojileri Dergisi Vol.18 Issue 2, pp.181-194

Belge Tipi:

Makale

Yayın Tarihi:

4/2025

Belge Dili:

Türkçe

Document ID (Org No.):

2001345766

pISSN:

1307-9697

X Social Media Platform

Spam Message

Machine Learning

Large Language Models

Computer VisionX Sosyal Medya Platformu

İstenmeyen Gönderi

Makine Öğrenmesi

Geniş Dil Modelleri

Bilgisayarlı Görme

Sayı

Makale detayları

Yazarlar: Ebutalha Camadan | Mehmet Şimşek

Yayıncı: Gazi University

Ana Başlıklar: Derin Öğrenme | Yarı ve Denetimsiz Öğrenme | Doğal Dil İşleme

Dergi: Bilişim Teknolojileri Dergisi Vol.18 Issue 2, pp.181-194

Belge Tipi: Makale

Yayın Tarihi: 4/2025

Belge Dili: Türkçe

Document ID (Org No.): 2001345766

pISSN: 1307-9697

Sayı

Yazarlar:

Ebutalha Camadan | Mehmet Şimşek

Yayıncı:

Gazi University

Ana Başlıklar:

Derin Öğrenme | Yarı ve Denetimsiz Öğrenme | Doğal Dil İşleme

Dergi:

Bilişim Teknolojileri Dergisi Vol.18 Issue 2, pp.181-194

Belge Tipi:

Makale

Yayın Tarihi:

4/2025

Belge Dili:

Türkçe

Document ID (Org No.):

2001345766

pISSN:

1307-9697