| With the rapid development of information technology,many security issues have become increasingly apparent,and information security has been elevated to a national strategy.Traditional information security is mainly realized around encryption technology and its application system,which can ensure the security of information content to a certain extent.However,with the development of data mining and machine learning technology in the era of big data,encrypted data,as a kind of abnormal data,is instead easy to expose the existence of secret information and become the key target of network data analysis.Compared with traditional encryption technology,steganography embeds secret messages in public information carriers to complete covert communication,thus not attracting the attention of third-party monitors,reducing the possibility of being attacked and detected,and effectively ensuring the security of information.Text is the most widely used information carrier in human daily life,and steganography methods using text as information hiding carrier have attracted the attention of many researchers,among which text-generated steganography methods have become a popular research direction in the field of information hiding because of their better steganography.At present,most of the generation-based text steganography methods only consider the probability distribution among text characters,and the subject of the generated steganographic text is uncontrollable,so there is still room for improvement in steganography.To address these specific problems and challenges,this paper deeply investigates the method and process of text steganography,and designs and proposes a text steganography method with better performance.The specific research work and innovations can be summarized as follows:(1)A graph embedding network is designed,which is capable of converting the topics and contents of the knowledge graph map into the corresponding graph vectors.Meanwhile,in order to solve the inherent sequential computation defects of recurrent neural networks,this paper adopts the Transformer architecture,which achieves efficient and parallel computation by invoking the self-attentive mechanism of global context modeling.(2)A topic-controlled generative steganography method is proposed,which can generate steganographic texts with specific topics.The data in the knowledge graph is introduced through the above graph embedding network to guide the generation process.The biggest advantage of introducing the knowledge graph is that there must be relationships among the entities in the knowledge graph,so the knowledge graph can provide data on relevant topics and contents,which ensures that the steganographic texts generated by this method have consistent topics.(3)Based on the generative text steganography method,we design and implement a steganographic communication system that transmits steganographic text containing secret steganography through a common channel,which verifies that the steganography method proposed in this paper has certain practical application value. |
