HGNN: A Hybrid Graph Neural Network Based on Transfer Learning for Linguistic Steganalysis

Merve Varol Arısoy

Makale detayları

HGNN: A Hybrid Graph Neural Network Based on Transfer Learning for Linguistic Steganalysis

Yazarlar:

Merve Varol Arısoy

Yayıncı:

Afyon Kocatepe University

Ana Başlıklar:

Bilgisayar Yazılımı | Yazılım Mühendisliği (Diğer)

Dergi:

Afyon Kocatepe Üniversitesi Fen Ve Mühendislik Bilimleri Dergisi Vol.24 Issue 5, pp.1138-1149

Belge Tipi:

Makale

Yayın Tarihi:

1/2024

Belge Dili:

İngilizce

Document ID (Org No.):

2001326900

pISSN:

2149-3367

Anahtar kelimeler:

Özet:

 Steganography, especially in the form of text generation based on secret messages, has become a current research topic. It is more difficult to identify the hidden message when it embedded directly into the text without using a cover text, and it also has a higher embedding capacity. Owing to the high rate of imperceptibility and resistance to steganalysis of this type steganography, it is essential that steganalysis methods, generate better performance. Although the complexity of deep learning models increases the accuracy rate, it also increases the inference time. In this study, a linguistic steganalysis was performed with a lower inference time and a higher accuracy rate. In the developed model, first, differences between non-stega and steganographic texts were modelled by a finetuned Bert using the custom dataset. The disparity information obtained by fine-tuned model was distilled into 3 separate networks, BertGCN, BertGAT and BertGIN, for faster and more accurate inference. Then, these 3 distilled networks were combined through Transfer Learning to form a new model. Experiments demonstrates that the proposed model surpass other methods in terms of the accuracy a success of 0.9879 at 3.22 bpw on text encoded through SAAC Encoding and the effectiveness of inference 1.09 second.

 Steganography, especially in the form of text generation based on secret messages, has become a current research topic. It is more difficult to identify the hidden message when it embedded directly into the text without using a cover text, and it also has a higher embedding capacity. Owing to the high rate of imperceptibility and resistance to steganalysis of this type steganography, it is essential that steganalysis methods, generate better performance. Although the complexity of deep learning models increases the accuracy rate, it also increases the inference time. In this study, a linguistic steganalysis was performed with a lower inference time and a higher accuracy rate. In the developed model, first, differences between non-stega and steganographic texts were modelled by a finetuned Bert using the custom dataset. The disparity information obtained by fine-tuned model was distilled into 3 separate networks, BertGCN, BertGAT and BertGIN, for faster and more accurate inference. Then, these 3 distilled networks were combined through Transfer Learning to form a new model. Experiments demonstrates that the proposed model surpass other methods in terms of the accuracy a success of 0.9879 at 3.22 bpw on text encoded through SAAC Encoding and the effectiveness of inference 1.09 second.

 Steganography, especially in the form of text generation based on secret messages, has become a current research topic. It is more difficult to identify the hidden message when it embedded directly into the text without using a cover text, and it also has a higher embedding capacity. Owing to the high rate of imperceptibility and resistance to steganalysis of this type steganography, it is essential that steganalysis methods, generate better performance. Although the complexity of deep learning models increases the accuracy rate, it also increases the inference time. In this study, a linguistic steganalysis was performed with a lower inference time and a higher accuracy rate. In the developed model, first, differences between non-stega and steganographic texts were modelled by a finetuned Bert using the custom dataset. The disparity information obtained by fine-tuned model was distilled into 3 separate networks, BertGCN, BertGAT and BertGIN, for faster and more accurate inference. Then, these 3 distilled networks were combined through Transfer Learning to form a new model. Experiments demonstrates that the proposed model surpass other methods in terms of the accuracy a success of 0.9879 at 3.22 bpw on text encoded through SAAC Encoding and the effectiveness of inference 1.09 second.

 Özellikle gizli mesajlara dayalı metin üretimi şeklindeki steganografi güncel bir araştırma konusu haline gelmiştir. Gizli mesajın kapak metni kullanılmadan doğrudan metnin içine gömülmesi durumunda tespit edilmesi daha zor olduğu gibi gömme kapasitesi de daha yüksektir. Bu tip steganografinin algılanamazlık oranının yüksek olması ve steganalize karşı direnci nedeniyle, steganaliz yöntemlerinin daha iyi performans üretmesi önemlidir. Derin öğrenme modellerinin karmaşıklığı doğruluk oranını arttırsa da çıkarım süresini de arttırmaktadır. Bu çalışmada, daha düşük çıkarım süresi ve daha yüksek doğruluk oranıyla dilsel steganaliz gerçekleştirilmiştir. Geliştirilen modelde öncelikle stega olmayan ve steganografik metinler arasındaki farklar, özel veri seti kullanılarak hassas ayarlı Bert tarafından modellendi. İnce ayarlı modelle elde edilen eşitsizlik bilgisi, daha hızlı ve daha doğru çıkarım için BertGCN, BertGAT ve BertGIN olmak üzere 3 ayrı ağa ayrıştırıldı. Daha sonra bu 3 damıtılmış ağ, Transfer Öğrenme yoluyla birleştirildi ve yeni bir model oluşturuldu. Deneyler, önerilen modelin doğruluk SAAC Kodlama yoluyla kodlanan metinde 3,22 bpwde 0,9879 başarı ve çıkarımın etkinliği 1,09 saniye açısından diğer yöntemleri geride bıraktığını göstermektedir.

Makale detayları

HGNN: A Hybrid Graph Neural Network Based on Transfer Learning for Linguistic Steganalysis

Yazarlar:

Merve Varol Arısoy

Yayıncı:

Afyon Kocatepe University

Ana Başlıklar:

Bilgisayar Yazılımı | Yazılım Mühendisliği (Diğer)

Dergi:

Afyon Kocatepe Üniversitesi Fen Ve Mühendislik Bilimleri Dergisi Vol.24 Issue 5, pp.1138-1149

Belge Tipi:

Makale

Yayın Tarihi:

1/2024

Belge Dili:

İngilizce

Document ID (Org No.):

2001326900

pISSN:

2149-3367

BERT

GAT (Graph Attention Network)

GCN (Graph Convolutional Networks)

GIN (Graph Isomorphism Network)

knowledge distillation

linguistic steganalysis

transfer learning.BERT

GAT (Graph Attention Network)

GCN (Graph Convolutional Networks)

GIN (Graph Isomorphism Network)

knowledge distillation

linguistic steganalysis

transfer learning.

Sayı

Makale detayları

Yazarlar: Merve Varol Arısoy

Yayıncı: Afyon Kocatepe University

Ana Başlıklar: Bilgisayar Yazılımı | Yazılım Mühendisliği (Diğer)

Dergi: Afyon Kocatepe Üniversitesi Fen Ve Mühendislik Bilimleri Dergisi Vol.24 Issue 5, pp.1138-1149

Belge Tipi: Makale

Yayın Tarihi: 1/2024

Belge Dili: İngilizce

Document ID (Org No.): 2001326900

pISSN: 2149-3367

Sayı

Yazarlar:

Merve Varol Arısoy

Yayıncı:

Afyon Kocatepe University

Ana Başlıklar:

Bilgisayar Yazılımı | Yazılım Mühendisliği (Diğer)

Dergi:

Afyon Kocatepe Üniversitesi Fen Ve Mühendislik Bilimleri Dergisi Vol.24 Issue 5, pp.1138-1149

Belge Tipi:

Makale

Yayın Tarihi:

1/2024

Belge Dili:

İngilizce

Document ID (Org No.):

2001326900

pISSN:

2149-3367