Unlocking Legal Insights: Applying Large Language Models to Analyze Divorce Judgments in China

Emma Zang, Yale University
Yishi Yin, Yale University
Zitong Wang, Chinese University of Hong Kong
Han Zhang, Princeton University

Legal judgments contain rich demographic-relevant information, yet their volume, complexity, and unstructured nature have made them an underutilized resource in demographic research. This study harnesses large language models (LLMs) to systematically analyze 3,602 Chinese divorce judgments from Zhejiang Province, drawn from a larger dataset of 72,102 cases nationwide between 2009 and 2016. Unlike traditional approaches reliant on manual coding or rule-based methods, fine-tuned LLMs, such as GPT-4o-mini, demonstrate superior accuracy in extracting key demographic variables—including gender, custody claims, and property division—by capturing implicit social cues and resolving ambiguities in legal text. Comparing ten computational strategies, we find that fine-tuned decoder-only models outperform traditional NLP methods, but performance also depends on factors such as hyperparameter tuning, preprocessing techniques, and linguistic context. This study is the first to apply LLMs to Chinese divorce judgments in demographic research, providing both empirical and methodological advancements. Our results highlight the scalability and efficiency of AI-driven text analysis, offering a robust alternative to resource-intensive manual coding. By systematically comparing rule-based methods, traditional machine learning, and transformer-based models, we offer a framework for selecting optimal classification techniques based on text structure and computational trade-offs. More broadly, our findings demonstrate how LLMs can expand the analytical scope of demographic research by processing vast unstructured datasets. While this study focuses on a sample of 3,602 cases, our method could efficiently code over 1.5 million legal judgments across China, paving the way for large-scale investigations of family law, socioeconomic disparities, and policy impacts. As AI continues to evolve, its integration into demographic research presents new opportunities to extract insights from legal texts at an unprecedented scale.

See extended abstract

 Presented in Session 110. Marriage and Divorce: The Role of Social Context and Legal Frameworks