Decoding Judicial Outcomes: Machine Learning Insights from Hong Kong Appeals

• In Event: Methods and Measures

Abstract

How can machine learning assist in classifying court decisions when coded information is unavailable, especially in nondemocratic court settings? This paper explores the application of unsupervised learning models, including Latent Dirichlet Allocation, word vector summaries, and the BERT model, in analyzing bilingual judgments. Leveraging a dataset of 25,000 appeal cases released by the Judiciary of Hong Kong, the study extracts document-level topic proportions and transfer-learning-based features. These features are then utilized to classify appeal outcomes, and the performance of different models is compared. The research also investigates the influence of judges' nationality on coded appeal outcomes, controlling for judgment text. By employing machine learning techniques, this paper demonstrates how political scientists can access previously unavailable information about nondemocratic courts, highlighting the potential impact of such approaches in enhancing our understanding of legal systems.

Author

• Siyun Jiang, IE University