Research On Text Steganography Based On Graph-to-Text Generation

At present,generation-based text steganography is a popular research area of information hiding.Taking text as a carrier and combining with technologies of text generation in natural language processing,a covert communication method with certain anti-detection ability can be built.However,in the process of researches and experiments,we can find that there are still some shortcomings in the existing methods.Nowadays,text steganography mostly adopts the method based on conditional probability coding.Under this method,the secret information hidden in the text tends to destroy the statistical distribution characteristics of the original text.This kind of method has the problem of the significant reduction of text quality when the embedding rate increases,and the topic of the generated steganographic texts is uncontrollable,so there is still room for improvement in concealment.In view of these specific problems and challenges,the methods and processes of text steganography are studied deeply,and a text steganography method with better performance is proposed in this paper.The specific research works and innovations can be summarized as follows:Firstly,this paper introduces a large knowledge graph and combines graph-to-text generation into text steganography.The model proposed in this paper can automatically generate the steganographic text with secret information from the knowledge graph,and different subgraphs in the knowledge graph contain different corresponding topics.The subgraphs are selected by choosing different topic words,and the steganographic text is further constructed by the graph-to-text generation model to ensure the topic controllability of the generated steganographic text.Secondly,a graph path coding method with corresponding detailed algorithm for the generation of steganographic text is proposed in this paper.Different from traditional text steganography methods,this method combines the structure of knowledge graph to encode the secret information in the process of graph path coding instead of using traditional conditional probability coding,which improves the quality of the generated steganography text without destroying the statistical distribution characteristics of the original text.Thirdly,combining the above parts,a topic controllable steganography method based on graph-to-text generation is proposed in this paper.The text steganography model is built,and the corresponding modules of topic matching,graph path coding,graph-to-text generation and secret information extraction are implemented.At the same time,different text steganography methods proposed during recent years are used for comparison in the aspects of semantic correlation,topic correlation,text quality and anti-detection ability.The experimental results show that the model proposed in this paper can improve the quality of the generated text,and significantly enhance the concealment of the steganographic text.