Fraud Web URL Detection Based on Bi-LSTM

Aiming at the problem of accurate identification of fraudulent web pages, a Bi-directional Long Short-Term Memory (Bi-LSTM) recognition model based on URL sequence mixed encoding of characters and words is proposed. According to the arrangement rules of URL sequences, special characters are used as segmentation points to be divided into different words, and then the URL sequences are encoded in two ways based on words and characters. The extraction results of the two encoding methods are added together. The extracted features are detected and classified using multiple classifiers such as Convolutional Neural Networks, Recurrent Neural Networks, and Bidirectional Long Short-Term Memory Networks. The experimental results show that the model using Bi-LSTM as the classifier can obtain the long-distance dependent features of URLs, and can achieve better classification results, and its accuracy can reach 99.63%.