BERT-BiLSTM-CRF-NER

Why we included this project

Named entity recognition is often the awkward step in Chinese-language text pipelines, and this project is one of the more complete BERT-based treatments of it. It couples Google's BERT embeddings with a BiLSTM-CRF sequence tagger, and the training script reads the standard token-per-line format and can export the finished model. There is also a Flask service that exposes NER and text classification over HTTP, which beats wrapping a notebook when you want it in production. The project is built around Chinese data, with Google BERT checkpoints and defaults that fit that use, though the README says other languages need only small code changes. The code is TensorFlow 1.x and has been dormant since early 2021, so treat it as a reference implementation and a template to adapt rather than something to drop into a modern stack unchanged.