Personality Privacy Protection Method of Social Users Based on Generative Adversarial Networks.

Area of Science:

• Computer Science
• Artificial Intelligence
• Cybersecurity

Background:

• Social engineering attacks exploit personality information for successful infiltration.
• Protecting personality privacy in text data is crucial for user security.
• Existing methods may not adequately balance privacy protection and semantic content preservation.

Purpose of the Study:

• To propose PerTransGAN, a generative adversarial network (GAN)-based method for text transformation.
• To obscure personality information within text data to mitigate social engineering risks.
• To preserve semantic similarity in transformed text while enhancing privacy.

Main Methods:

• Utilizing generative adversarial networks (GANs) with reinforcement learning for text transformation.
• Employing a discriminator's output as a reward signal to train the generator.
• Incorporating semantic guidance signals and a penalty item in the generator's loss function.
• Designing semantic and personality modules to ensure content retention and privacy.

Main Results:

• PerTransGAN improved content retention by 0.11 and achieved the highest BLEU score compared to baseline models.
• The addition of penalty and personality modules reduced personality classifier accuracy on generated text by 20% compared to original data.
• The self-attention and semantic modules enhanced text content retention.

Conclusions:

• PerTransGAN effectively preserves user personality privacy in text data through transformation.
• The method successfully maintains semantic similarity while hindering privacy theft by attackers.
• PerTransGAN offers a robust solution for protecting sensitive personality information in digital communications.