The Research And Implementation On URL Detection And Generation Technology Based On Deep Learning

In recent years,network security attacks become more and more frequent.Among them,malicious URL plays an important role in network attacks,such as phishing sites,identity theft,financial fraud,malicious software,spam and so on.How to effectively detect malicious URL has become a key problem to prevent the increasing network attacks.The traditional methods of malicious URL detection have been unable to keep up with the rapid growth of malicious URL because they need a lot of feature recognition work.It has become a research problem in the field of URL detection to construct a generalizable model based on deep learning technology to realize URL recognition and classification.The main research contents of this thesis are as follows:1.A new malicious URL recognition model based on URL imaging technology is proposed.In this model,the one-dimensional URL word vector is converted into a twodimensional image by Gram Angle Field graphic coding method,and the converted twodimensional image is used to mine out more inherent features of the URL to improve the classification accuracy.Then,a URL binary classification model “Binary-Inception CNN”is designed.This model uses the convolution neural network to perform multiple convolution and pooling operations concurrently on the input GAF images,and enhances network adaptability by using a series of convolution kernels in different sizes.Thus more abundant multi-scale image characteristics can be extracted,and the computational cost is lower.Finally,a series of experiments are carried out on a large URL dataset,and the experimental results show that the proposed model is superior to the current mainstream malicious URL recognition methods in many indicators.2.A new URL generation and multi-classification model based on Generation Adversarial Network(GAN)is proposed.In order to solve the problem of the imbalance number of different types of URLs in the real network environment.This model designs an encoder to encode URLs firstly,then trains a GAN network for each type of URL by using a small amount of the same type sample data.The GAN networks can generate a large number of synthetic URLs close to the real sample.Finally,a URL multiple classification model “Multi-Inception CNN” is proposed to realize the automatic classification task of different types of URLs.Through the similarity analysis of the original URLs and the generated URLs,the GAN network in this model can generate the synthesized URL data which is close to the real sample.Meanwhile,a large number of experimental results also show that the model can achieve a better effect of URL multiclassification under the condition of less sample data.3.A malicious URL detection system URLGuard is designed and implemented.The system integrates three core models of binary-inception CNN,GAN network and multiinception CNN to realize malicious URL detection,different types of URL generation and multiple URL classification.Through the function and performance test of the system,this thesis verifies that the URLGuard system has good availability and robustness.