Key Entity Extraction-Enhanced Semantic Communication Systems over Cloud-Edge-Device Architecture

Semantic communications have shown strong robustness in bandwidth-limited and low signal-to-noise ratio (low-SNR) channels by transmitting task-aware semantics rather than raw data. However, most existing systems neglect the heterogeneity of semantic information across words, leading to higher semantic distortion when errors occur in high-information words. To address this, we propose a key entity extraction-enhanced semantic communication system, named KEE-SC, within a cloud-edge-device collaborative architecture. The system integrates key entity extraction with a semantic diversity optimization algorithm to generate high-quality and diverse key entities, and employs a multi-layer cross-attention mechanism to improve semantic separation and fusion for short-long text interactions. By jointly transmitting shared key entity vectors and residual information, KEE-SC significantly reduces the number of transmitted symbols while maintaining semantic fidelity. Simulation results show that the proposed system outperforms existing baselines, particularly under low-SNR conditions.