Linguistic Steganalysis via Fusing Multi-Granularity Attentional Text Features

Deep learning based language models have improved generation-based linguistic steganography, posing a huge challenge for steganalysis. The existing neural-network-based steganalysis methods are incompetent to deal with complicated text because they only extract single-granularity features such as global or local features. To fuse multi-granularity features, we present novel method on attentional bidirectional long-short-term-memory (BiLSTM) and short-cut dense convolutional neural network (CNN). BiLSTM equipped the scaled dot-product attention mechanism is used capture long dependency representations of input sentence. CNN connection exploited sufficient semantic from word embedding matrix. We connect two structures in parallel, concatenate classify stego cover texts. results comparative experiments demonstrate that proposed superior state-of-the-art