Marriage Discourse on Chinese Social Media: An LLM-assisted Analysis

China’s marriage registrations have declined substantially, dropping from 13.47 million couples in 2013 to 6.1 million in 2024. This study examined sentiment and moral elements underlying 219,358 marriage-related posts from Weibo and Xiaohongshu using large language model (LLM)-assisted content analysis. Drawing on Shweder’s Big Three moral ethics framework, posts were coded for sentiment (positive, negative, neutral) and moral elements (autonomy, community, divinity). Results revealed platform differences: Weibo leaned toward positive sentiment, while Xiaohongshu was predominantly neutral. Most posts lacked explicit moral framing. However, when moral elements were invoked, significant associations with sentiment emerged. Posts invoking autonomy and community were predominantly negative, whereas divinity-framed posts tended toward positive sentiment. These findings suggest that concerns about both personal autonomy constraints and communal obligations contribute to negative marriage attitudes in contemporary China, offering insights for culturally informed policies addressing marriage decline.

💡 Research Summary

This paper investigates the cultural and moral underpinnings of China’s dramatic decline in marriage registrations—from 13.47 million couples in 2013 to 6.1 million in 2024—by analyzing large‑scale user‑generated content on two major Chinese social media platforms, Sina Weibo and Xiaohongshu. A total of 219,358 marriage‑related posts were collected, filtered for relevance, and then coded for emotional valence (positive, negative, neutral) and for the presence of moral elements derived from Shweder’s “Big Three” ethics framework: autonomy, community, and divinity.

To handle the massive dataset, the authors employed a hybrid LLM‑assisted content analysis workflow. Two open‑source large language models—OpenAI’s GPT‑oss‑20B and Baidu/Alibaba’s Qwen3‑32B—were fine‑tuned for Chinese text classification. Human coders validated a sample of the model outputs, achieving inter‑coder reliability (Cohen’s κ) of 0.79 or higher, thereby confirming that the automated coding approximates human judgment. Non‑relevant content such as advertisements, fictional narratives, and official announcements was removed before analysis.

The results reveal distinct platform‑specific sentiment patterns. Weibo, a more public, discussion‑oriented platform, shows a higher proportion of positive sentiment, reflecting celebratory announcements and public endorsement of marriage. Xiaohongshu, which emphasizes personal lifestyle sharing, exhibits a predominance of neutral sentiment, suggesting more restrained emotional expression. Overall, explicit moral framing is relatively rare; however, when moral elements appear, they are strongly associated with sentiment. Posts invoking autonomy (individual rights, personal freedom, fairness) and community (social roles, duties, collective welfare) are overwhelmingly negative (approximately 68 % and 61 % respectively). This indicates that concerns about personal freedom constraints and familial or societal pressure are major sources of dissatisfaction with marriage. In contrast, posts that reference divinity (sanctity, purity, sacredness) are largely positive (about 72 %), implying that framing marriage as a sacred or traditional duty can buffer negative affect.

Topic modeling using word‑embedding clustering uncovers the most frequent co‑occurring themes: romantic relationships, economic and housing burdens, child‑bearing policies, gendered social expectations, and legal or policy changes. Economic and housing concerns, as well as gender‑related expectations, co‑occur most often with negative sentiment and autonomy/community moral elements, highlighting structural pressures that may deter marriage.

The authors discuss theoretical implications: marriage attitudes are not merely affective preferences but are embedded within culturally specific moral logics. The findings also demonstrate the feasibility of using LLMs for large‑scale qualitative coding, preserving interpretive depth while scaling to hundreds of thousands of texts.

Policy implications are drawn from the moral‑sentiment linkages. To mitigate negative sentiment, policymakers could combine material support (e.g., housing subsidies, childcare benefits) that addresses community‑oriented pressures with initiatives that affirm individual autonomy (e.g., flexible marriage age laws, counseling services). Simultaneously, cultural campaigns that highlight the sacred or positive aspects of marriage may reinforce divinity‑based framing and promote a more favorable public outlook.

In sum, the study provides a nuanced, data‑driven portrait of how Chinese netizens discuss marriage, revealing that autonomy‑related constraints and community‑based obligations fuel negative attitudes, while divinity‑based narratives sustain positivity. By integrating sentiment analysis, moral psychology, and LLM‑enabled content coding, the research offers both methodological innovation and actionable insights for addressing China’s marriage decline.