blmoistawinde/HarvestText

Combines unsupervised statistical methods (TextRank, TextTiling, new word discovery rules) with domain knowledge integration to enable weak-supervision NLP tasks on Chinese text, particularly effective for entity disambiguation and relationship extraction from noisy sources like social media and novels. Integrates PyHanLP for dependency parsing and NER, supports modular sentiment analysis through seed word bootstrapping, and provides built-in domain vocabularies (legal, financial, culinary, IT) for immediate application without extensive labeled data.

2,603 stars and 1,333 monthly downloads. Used by 1 other package. No commits in the last 6 months. Available on PyPI.