Gaokao News Frame Analysis Based on Natural Language Processing Models
Abstract
Gaokao, a major public issue characterized by its selective, institutional, and symbolic nature, has long stood at the intersection of national politics and public discourse in China. Media narratives surrounding the Gaokao are not merely journalistic reproductions but also critical arenas for ideological dissemination and social cognition construction.
This study adopts natural language processing (NLP) as its core methodology to systematically analyze the news frames and emotional expressions embedded in Chinese media reports on the Gaokao from 1977 to 2025. To overcome the limitations of previous research relying on static corpora and single-event analysis, a comprehensive two-tier classification system of general and issue-specific frames was developed. Advanced pre-trained language models (BERT and DeepSeek) were applied for large-scale frame detection, complemented by the CARER model to perform emotion modeling and establish an integrated framework for simultaneous frame and emotion analysis.
Findings indicate that media ownership structure and regional background significantly shaped news frame selection and emotional expression: state-owned media predominantly focused on policy legitimacy and political stability, while market-oriented and coastal media emphasized social ethics, demands for fairness, and heightened emotional tension.
Furthermore, the proposed Frame-Emotion Dual-Track Interaction Model attempts to explain how media collaboratively shape public understanding and value judgments of the education system through the interplay between news framing and emotional modulation. This research addresses a critical gap in current news framing studies by integrating automated annotation, longitudinal analysis, and emotion modeling, providing a novel methodological approach to understanding the mediated transmission of policy issues in the media landscape.