The existing neural-network-based linguistic steganalysis methods are incompetent to deal with complicated text because they only extract single-granularity features such as global or local text ...