keras2_crf | Models based on LSTM and Conditional Random Fields

keras2_crf is a Jupyter Notebook library. keras2_crf has no bugs, it has no vulnerabilities and it has low support. You can download it from GitHub.

Models based on LSTM and Conditional Random Fields (CRF) can obtain State of the Art results on sequence labelling tasks in Natual Language Processing, such as Named Entity Recognition (NER), Part of Speech (POS) and Chunking [1, 2]. Despite of this, there is no official implementation of a CRF layer, more specifically Chain CRF, available on Keras. Currently there's one implementation on keras-contrib (example of usage here) by linxihui and another one kindly provided by phipleg (an extensive merge discussion thread can be found here). Unfortunatelly the latter implementation doesn't work on the most recent versions of Keras 2, although such limitation can be easily overcome by modifying just two lines of code, as shown by Hironsan on this commit. The goal of this repository consists in showing which changes have to be made in the code kindly developed by phipleg and contributors, so it can work on the most recent version of Keras (currently 2.1.5); and to provide a minimum working example of a CRF-based model for POS tagging. The present implementation is inspired on the code developed by nreimers, emnlp2017-bilstm-cnn-crf [3].