The doctrine of the "best interests of the child" is the primary consideration for family court judges in making post-divorce custody decisions. However, a precise understanding of the "best interests" standard remains elusive. As Article 1055-1 of Civil Code requires judges in Taiwan to consider factors such as the age, gender, health condition and wish of the child, the character, occupation, financial condition and life style of the parents, the judge still has a wide range of discretion to reach the final decision. This article therefore aims to clarify: among the key factors, which ones take top priority in child custody decisions? Do judges in Taiwan generally favor father or mother after considering these factors? And when facts are properly gathered, is it possible to predict judicial decisions by computer algorithms? To answer the questions, over two thousands child custody cases are examined and features are labeled, then natural language processing (NLP) is employed to classify and analyze these decisions to make reliability assessments. Although this article is a work in progress, it discusses the preliminary results of an ongoing four-year research project supported by Taiwan Ministry of Technology. It is expected that with the help of NLP and descriptive statistics, the findings of this research will help to deliver a more predictable outcome for custody disputes. Moreover, the established judicial database can be a useful source toward a better understanding of judicial behavior.
Keywords: best interests of the child, child custody, empirical legal studies, judicial behavior, judicial decision prediction