Research On Phishing Website Detection Technology In Dual-structural Network

In recent years,with the rapid development of the Internet,various types of information resources have been circulated on the Internet.While bringing great convenience,various security issues such as phishing,cybercrime,and privacy leakage have become even more prominent.The dual-structural network argues integrating the current Internet architecture with a broadcast-storage secondary structural network characterized by “radiation-copy” model,to realize the "de-redundancy" of the current Internet with the innovative ideas of physical transformation and dual structure.At the same time,UCL(Uniform Content Label)is used to provide network security related services such as phishing website detection at the user terminal to protect user data security.However,the traditional phishing website detection methods rely heavily on phishing blacklist,and the untimely update of the blacklist will lead to a rapid increase in the false negative rate;traditional phishing website detection method based on machine learning need to extract numerous features,resulting in slow detection speed and poor accuracy.The users visit a large number of websites every day in the dual-structural network.They have high requirements for accuracy,false negative rate,false positive rate,and detection speed.How to minimize false negative rate,false positive rate,increase accuracy and the rapid detection of phishing websites is the face of traditional phishing website detection in the dual-structural network.Aiming at the requirements and characteristics of the dual-structural network,this dissertation proposes a phishing URL detection algorithm based on convolutional neural network and long-short term memory network,abbreviated as CNN-LSTM-PUDA.On this basis,a phishing website detection system for dual-structural network(PWD-DSN)is designed to help users detect phishing websites in real time,efficiently and accurately.The specific work of the dissertation is as follows:(1)Aiming at the problem of low accuracy of the traditional phishing detection methods based on URL characteristics,a phishing URL detection algorithm CNN-LSTM-PUDA,which is suitable for dual-structural network,is proposed.Firstly,consider the URL as a string sequence,then encode the URL into an onehot two-dimensional sparse matrix,followed by convert it into a dense character embedding matrix.Secondly,input it into the convolutional neural network to extract the local depth feature.Thirdly,entered into long-short term memory network to capture before and after associations of the URL.Finally,access the softmax model to classify the URL.The algorithm avoids complex feature engineering,extracts the local correlation features through convolutional neural network,and learns long-term dependence through long-short term memory network.The phishing website can be detected quickly and relatively accurately.(2)In order to satisfy the urgent need of end-users to detect phishing websites accurately and rapidly in dual-structural network.Based on the characteristics of dual-structural network,a phishing website detection system named PWD-DSN is designed based on the CNN-LSTM-PUDA algorithm.First of all,a storage detection and update strategy is designed for phishing blacklist and whitelist;then in order to further increase the accuracy of phishing detection,reduce the rate of false negative and false positive,a multi-features fusion method combining deep URL feature(MFFCDUF)is proposed.The algorithm firstly extracts URL features,source features of website,and text features of website.Combining the classification result of CNN-LSTM-PUDA algorithm,the four features are merged,and then XGBoost is used to classify the fusion features.Finally,in order to speed up the detection of phishing websites,improving the softmax model classification results,a two level detection algorithm(TLDA)for phishing websites detection in dual-structure network is proposed.(3)Based on the dual-structural network prototype system,a phishing website detection prototype system is designed and implemented based on the above algorithms.In the system,real data sets were used to verify the CNN-LSTM-PUDA,MFFCDUF and TLDA algorithms proposed in this dissertation.The experimental results show that,compared to the traditional phishing website URL detection algorithm,the CNN-LSTM-PUDA is more accurate,and the MFFCDUF feature fusion method can further improve the accuracy of phishing website detection;TLDA can significantly reduce the detection time while ensuring the accuracy.