Research On Network Sensitive Information Perception Based On Deep Learning

The scope of network information service has been expanded gradually,showing obvious characteristics of intelligence and refinement.Frequent Internet behavior produces a large amount of information,resulting in a large amount of network information is difficult to deal with;With the characteristics of openness,strong interaction,high concealment and fast propagation,the cyberspace has become the main battlefield for criminals to carry out activities endangering public security,which puts forward higher requirements for the national security department's action strategies and means in the network governance module.The thesis focuses on the types of text information in mass network information,proposes a method and model of network sensitive information perception based on deep learning method,and carries out feature extraction and Chinese named entity recognition respectively for network text information.Sensitivity in multifarious network information,there are many text messages,text information itself has the polysemy,ambiguity,the problems such as unclear,in the process of text analysis to deal with huge noise,combined with the network space information in fast speed,the intelligence analysis department can't get or intercept information quickly,also it is difficult to get the analysis a large number of valuable information,available.Although the big data technology is becoming more and more mature and has been successfully applied in all walks of life,the big data technology is obviously unable to meet the requirements of high efficiency and accuracy of data processing for large-scale and messy text data.Therefore,the deep learning method is selected to improve the efficiency and obtain the deep information more comprehensively.Firstly,the sparse network essay this feature extracting,since the encoder using sparse active learning process of the encoder,the target passage by high dimensional vector into a low dimensional vector,eliminate distractions,to ensure that the low dimensional vector contains the essential features of the original data,and to verify the effectiveness of the method,set up the traditional feature extraction method for comparison,and verified with the experiment;Secondly,the thesis selects the length of the two-way memory networks and the conditional random field with the way of combining Chinese text of named entity recognition,set length only use two-way memory networks for experiment of Chinese named entity recognition as a control group,both high accuracy and effectiveness of deep learning method has been verified,thus confirmed in this thesis,the proposed method and model which can realize the network active rapid sensitive information of perception.In network information in a wide range of clutter,the sensitivity of text information concerns and defines is an important task,and involves the accurate extraction of text characteristic,the sensitivity definition,sensitive to a specific entity recognition and so on,only the depth of mining characteristic properties of text information potential,can accurate perception and found that it may have sensitivity,thus for maintenance of the network environment and public social security early warning to assist in advance.