Identifikasi Pesan Teks Berunsur Phishing Menggunakan Kombinasi Model Indobert Embedding dan Algoritma LSTM

Abstract

Phishing is an illegal method of obtaining information such as usernames, passwords, and credit card information by pretending to be a legitimate electronic communication company. The factors that lead to the threat of phishing attacks are the lack of user knowledge, psychology, privacy of users' social networking services, and many are trapped by phishing text messages that resemble official communication language. Furthermore, the main problem in detecting phishing text messages is the limitation of previous traditional detection methods that only use rule-based detection and keyword matching, and are static and English-focused. As a result, it is difficult to handle new variations of phishing patterns and those that understand Indonesian. This is the basis for the need for an effective approach to detect phishing text messages either via SMS, e-mail, or other sources. In this study, researchers combined the LSTM and IndoBERT Embedding algorithms to detect phishing text messages in Indonesian. This research uses 9867 data in training and evaluating the model. The results showed that the model and system built with the combination of LSTM and IndoBERT Embedding algorithms were able to detect phishing with an accuracy of 97.93%.