<![CDATA[General encryption method provides a low level of security against brute-force attacks, which attempt every possible key through the decryption process. The Honey Encryption (HE) method was developed to overcome these problems. However, the latest development of HE implemented in chat conversations results in conversations that do not correlate between one chat and another when the ciphertext is decrypted using the wrong key (decoy message). The absence of context relation in the decoy message results in the decryption results of the brute force attack becoming easy to analyze, which can make it possible for the adversary to obtain the correct decryption results and map the keys used. This study aims to improve the context relation or contextual naturalness between chats in HE decoy messages. This improvement is intended to generate contextually natural sentences such that an attacker cannot distinguish between messages decrypted using the correct key and those decrypted using the incorrect key. This process reduces the likelihood of success during a brute-force attack for obtaining the plaintext and the key.. A contextual naturalness decoy message is generated by utilizing keyword detection (KeyBERT) and keyword alternative mapping as key points for context generation. Additionally, the generated keywords are used in sentence generation with a Masked Language Model (in this study, Electra) for generating word candidates. The implementation of KeyBERT and Electra in this study shows that 22 out of 32 decoy chat messages can be categorized as natural conversations. Finally, it can be concluded that the proposed method has contextual relation between decoy chat messages.]]>
Copyrights © 2026
