JioNLP

Why we included this project

Anyone building software that handles Chinese text probably knows the pain of rewriting the same cleanup code on every project. JioNLP collects those utilities in one place: it strips noise such as HTML and URLs from raw text, parses Chinese time expressions into concrete timestamps and durations, and extracts money amounts, phone numbers, ID cards, and addresses. The time parser and the address and ID-card extractors are the parts that earn their keep, since Chinese dates and administrative divisions have enough edge cases that doing them from scratch is a lot of fiddly work. The library also converts between simplified and traditional Chinese, generates pinyin, and offers a few data-augmentation tricks like back-translation and homophone substitution when you need more training examples. It installs as a normal Python package and runs without heavy model dependencies, so it drops into an existing NLP stack without much ceremony.