Double Graph Convolution Network with Knowledge Distillation for International Media Portrait Analysis of COVID-19

Global media with international influence play a crucial role in shaping international public opinion related to China. These media report on objective events and shape people’s perceptions and viewpoints. So, we are studying how international public opinion forms are crucial for constructing a positive national image, improving the international public opinion environment, and resisting various forms of ideological infiltration. This paper focuses on conducting an international media portrait analysis of China during the COVID-19 pandemic as a typical case study. The task involves fine-grained sentiment classification through aspect-based sentiment analysis. Due to the complexity of online comments and inaccurate parsing results, improving the accuracy of this task presents significant challenges. We propose a KD-Dual-GCN model based on knowledge distillation and dual graph convolutional networks to address these challenges, considering both syntactic structure and semantic correlation. Our SynGCN module leverages rich syntactic knowledge to mitigate dependency parsing errors, and our SemGCN module incorporates a self-attention mechanism to capture semantic correlation. Additionally, we employ knowledge distillation to reduce model latency and network parameters. The results from experiments conducted on three standard datasets show that our KD-Dual-GCN model outperforms existing methods, validating the effectiveness of our proposed approach.