Research On Malicious URLs Detection Based On Neural Network Model

With the rapid development of the Internet,attacks related to network security are also increasing.The variety of network security attacks has affected our daily life to a certain extent.As the scale of cybercrime continues to expand and has already caused a certain scale of economic loss,it is very important to protect potential victims from various attacks.Among them,malicious URLs are commonly used by intruders as a network security intrusion method.When ordinary users visit malicious URLs,they do not realize that they are malicious websites,which often leaves opportunities for attackers.In the current malicious URL detection system,there is still a problem that the detection accuracy is not high enough for the imbalanced data set categories,and the detection accuracy is still not high enough for specific categories of malicious URL attacks.Therefore,it is necessary to conduct research on malicious URL attack detection.This thesis mainly focuses on the problems existing in the detection of malicious URLs,and conducts the following researches:(1)In view of the problems existing in information security and network security in the field of malicious URL detection,the current domestic and foreign research achievements in the field of URL attacks are described in detail,the types of malicious URL attacks are introduced in detail,and the main current Threats in economic and political fields,detailed introduction and derivation of various basic networks in neural networks,and two effective representations for URL features(2)Aiming at the problem that the detection accuracy of malicious URLs is not high enough due to the problem of unbalanced samples,a multi-classification detection method of malicious URLs based on FTCNN-Bi LSTM is proposed.The FTCNN-Bi LSTM hybrid model is a deep network model composed of a convolutional neural network and a bidirectional long short-term memory neural network.The vectorized URL features are learned through a convolutional neural network,and the features are further learned through a bidirectional long-term and short-term memory neural network.The network model structure is optimized using directional random deactivation regularization,and the model training is corrected using a focus loss function.Simulation experiments show that the FTCNN-Bi LSTM model has better detection performance than other models.(3)In order to solve the problem that the detection accuracy of specific types of malicious URL attacks is not good enough,a malicious URL detection model based on DCNN-GRU-Att is proposed.The effective features in the URL are extracted by dilated convolution,and the features are further screened by the gated recurrent unit neural network,and the neural network structure is optimized by combining the attention mechanism to improve the learning ability of the model.At the same time,the simulation experiments are carried out,and the results show that DCNN-GRU-Att can effectively improve the detection accuracy of malicious URL attacks.This thesis mainly studies the field of malicious URL attack detection,and proposes the following innovations:(1)A new malicious URL detection method FTCNN-Bi LSTM is proposed,which uses convolutional neural network and bidirectional long short-term memory neural network to fully learn malicious URL features,combines the advantages of the two neural networks,uses directional random deactivation regularization and focused loss function optimization Model structure,thereby improving the detection accuracy of the model and enhancing the robustness of the model.(2)Combining the characteristics of dilated convolution and attention mechanism,a malicious URL detection method based on DCNN-GRU-Att is proposed,which can effectively improve the problem between model complexity and URL feature expression ability,and extract effective key malicious URLs.URL features,thereby improving the ability of the model to learn URL features from the input space and improving the accuracy of malicious URL model detection.