MalText: Finding Malicious Account in Social Network using Novel Regulizer in Dynamic CNN

Abstract

Abstract Online Social Networks (OSN) are well-known platforms for exchanging various information. However, one of the most critical OSN obstacles is malicious accounts. The attacker harnesses malicious accounts in the infected system to spread false information, such as malware, viruses, and harmful URLs. Based on the significant achievement of the CNN model in various fields, we propose a dynamic CNN using a novel regulizer to handle malicious account classification with user comments as features. Using the proposed regulizer, we obtain higher scores with a testing accuracy of 0.9948 and a testing loss of 0.0984 using unseen comment features. Our experimental results demonstrate that the proposed model can significantly improve the classifier's performance by producing high accuracy with minimal loss. Therefore, the proposed method can be a promising solution for analyzing large-scale user text datasets to classify malicious text in practical implementation.