Amid rising global divorce rates, this study uses large language models (LLMs) to analyze over 1.1 million first-instance divorce judgments from Court Judgement dataset. The project pursues three core questions: (1) identifying major divorce motives and their evolution over 40 years; (2) linking motives to couple-level dyadic traits (e.g., age, education, assortative mating); and (3) examining how macro-level factors—regional GDP, urbanization, gender ideology—shape marital dissolution. By combining computational text analysis with socio-economic contextual data, it offers scalable, nuanced insights into intimate relationship dynamics and broader structural changes.