sequence-labeling-BiLSTM-CRF

Why we included this project

This is a clean TensorFlow implementation of the BiLSTM+CRF model, the standard neural approach to token-level sequence labeling since Lample et al. proposed it in 2016. It covers the usual tasks in this area, named entity recognition, part-of-speech tagging, chunking, and similar problems, and lets you swap encoders, decoders, and labeling schemes like BIO or BIESO through a single config file. The code is split into clear modules, which makes it a good way to see how the pieces of a sequence labeling system fit together. It also includes training, testing, and interactive prediction modes, plus a small Django web app for trying results in a browser. One caveat: it targets TensorFlow 1.x, so treat it as a learning and prototyping reference rather than something to drop into a modern production stack.