Introduction
In the People’s Republic of China (PRC), state-run media outlets like the People’s Daily function as critical barometers of party ideology and internal power dynamics. While historians have qualitatively observed that the portrayal of political figures shifts in response to "political tides," systematically quantifying these narrative transitions across decades of archival text remains a significant challenge. This study introduces a quantitative framework to analyze how these newspapers strategically adjust their narratives to align with evolving political climates.
Methodology
The research utilizes a multi-pronged computational approach to process over 50 years of newspaper archives:
- Descriptive Statistics & Sentiment Analysis: Frequency of name mentions and sentiment scores (via Google Cloud Natural Language API) were tracked for key political figures.
- Change Point Detection (CPD): Kernel-based CPD was applied to identify statistically significant "turning points" in mention frequency and sentiment, correlating them with historical events.
- Topic Modeling (LDA): Latent Dirichlet Allocation extracted dominant themes (e.g., "Revolution" vs. "Economy") to map the evolution of framing across decades.
- Masked Language Modeling (MLM): A pre-trained Chinese BERT model was used to predict redacted names, measuring the "uniqueness" of the rhetorical style associated with specific leaders.
The primary datasets included ~1.3 million articles from the People’s Daily (1946–2003) and ~570,000 articles from Reference News (1957–2002).
Key Insights: The Rhetorical Fingerprint of Power
A central finding of the study is the "Posthumous Respect" trend: for most politicians, sentiment scores became significantly more positive following their death, reflecting a journalistic tendency toward reverence for deceased figures.
The Mao Exception
Mao Zedong represents a unique departure from this trend. Unlike his peers, Mao’s sentiment scores were highest while he was alive, particularly peaking during the Cultural Revolution. Posthumously, the tone shifted toward a more neutral, "objective" stance.
As shown in the time-series data, the frequency of Mao's mentions aligns closely with his political dominance, reaching a zenith between 1966 and 1971. This period also corresponds with a peak in positive sentiment, illustrating how state rhetoric reinforced his status during the height of his power.
Framing and Language Uniqueness
The analysis revealed a clear ideological shift in the People’s Daily, moving from themes of "Production" and "Revolution" in the mid-20th century to "Economy" and "Development" by the 1980s. Furthermore, the BERT model demonstrated that the language used to describe top-tier leaders like Mao and Zhou Enlai was highly individualized. Mao, in particular, had a "Uniqueness Ratio" of 8.08, indicating that the surrounding political rhetoric was so specific to him that the AI could identify him with high accuracy even without his name.
Conclusion
This study demonstrates that the "rhetorical fingerprints" of Chinese politicians are inextricably linked to their power status and the prevailing party ideology. By identifying mathematical "turning points" that align with historical shifts—such as the end of the Cultural Revolution or the death of a leader—the authors provide a robust, data-driven method for historical analysis. Future work may leverage Large Language Models (LLMs) to further refine framing analysis and explore complex power networks within state-run media.
Summary updated