P-BERT: Polished Up Bidirectional Encoder Representations from Transformers for Predicting Malicious URL to Preserve Privacy

The usage of internet among user made online social network (OSN) like twitter, facebook, weibo to become popular. Users share their thoughts and perspective on aspects on OSN. In OSN the biggest security threat is the malicious Uniform Resource Locator (URLs) to prevent from privacy. Researchers have found few methods to detect the malicious URL by hard coded eminent features, block listing the URLs. These methods have limitations such as not all malicious URLs are blacklisted and many important features are not considered in hard coding method. Evolution of deep learning techniques have made to extract and analyses the features by own and solutions can be derived easily. In this paper, a novel feature engineering approach and polished up Bidirectional Encoder Representations from Transformers (BERT) is proposed to comprehensively detect the malicious Uniform Resource Locator (URLs). The results show that proposed model gives 98.79% of overall accuracy is achieved which out performs from the state of art models.