Part-of-speech-Enhanced Text Semantic Communication Framework: A GCN-Based Gate Mechanism

In this paper, we propose a part-of-speech-enhanced text semantic communication (PSTSC) framework to enhance the robustness of communication systems. We have established a multi-encoder Transformer architecture that enhances semantic information transmission by incorporating part-of-speech (POS) tagging and a gate mechanism. Specifically, at the transmitter, POS tagging is introduced into the embedding layer of the semantic encoder, using contextual knowledge to enhance the encoder’s performance. At the receiver, we design a gate mechanism based on graph convolutional neural (GCN) networks that dynamically adjust the weights of the input signal according to semantic importance, thereby enhancing the noise resistance of the received signal and increasing system transmission reliability. Furthermore, we validate the effectiveness of each component of the model through ablation experiments. Simulation results demonstrate that the PSTSC framework can increase the BLEU score and sentence similarity by about 48% and 23% compared with the existing schemes, has significant performance advantages under low signal-to-noise ratio (SNR) conditions, and effectively overcomes the cliff effect.