insuranceqa-corpus-zh

Why we included this project

This repository offers a Chinese translation of the InsuranceQA dataset, pairing questions asked by real users with answers written by insurance professionals, in both Chinese and English. That origin matters for training: the phrasing is genuinely how people ask, not synthetic examples. The corpus comes in two forms, raw translated question-answer text and cleaned, tokenized pairs with labels that are ready for supervised tasks like answer selection. A bundled Python package loads the train, validation, and test splits directly, which suits teams working on insurance chatbots, FAQ retrieval, or customer support NLP. The main hurdle is access: downloading the data requires a license certificate from the maintainer, so factor that step into your evaluation.