425776024/nlpcda

Provides nine distinct augmentation strategies for Chinese NLP—including entity/synonym/homophone replacement, character deletion and transposition, and generative methods via SimBERT—while preserving semantic meaning through targeted filtering (e.g., dates/numbers remain unchanged). Offers specialized support for NER tasks in BIO format, back-translation augmentation via Baidu/Google APIs, and integrates custom lexicon injection via jieba tokenizer. Designed to improve model generalization and robustness across classification, NER, and retrieval tasks without sacrificing label integrity.

1,878 stars and 405 monthly downloads. Used by 1 other package. No commits in the last 6 months. Available on PyPI.