Smart vehicles networks: BERT self-attention mechanisms for cyber-physical system security

People can freely communicate on a travel-related subject over the internet through IoT on smart vehicles to maintain their interpersonal relationships. Cyber security challenges are brought on by the rising use of smart vehicle communications and information. Attackers use the chance to fabricate communications to obtain users’ personal information from smart vehicles. It is a significant duty to distinguish genuine postings from phishing messages and violence among the millions of messages on smart vehicle networks. The Bidirectional Encoder Representations from Transformers (BERT) method with self-attention is proposed in this paper to process textual threats, vulnerability data, and cyber-attacks. The BERT-SA model is fine-tuned to enhance performance. The self-attention method of a proposed model determines the contextual relationships among words in a text input, i.e., governs the relative reputation of each word in the phrase by examining its exact placement inside the sentence. This model’s primary goal is to accurately carry out tasks specialized in cyber-security. The BERT model performs better than other neural networks models like CNN, RNN, and LSTM. Sentiment140, T4SA datasets are used to assess the effectiveness of the BERT model. The experimental results show that the proposed model enhances the accuracy by up to 96.65%.