This is a general Natural Language Processing (NLP) system that comprised of a Named Entity Recognition (NER) module and a Relation Extraction (RE) module. The Information Extraction Document (IE) class is the main data structure used through out the training, evaluation, and prediction for both NER and RE.
Main frameworks: PyTorch, Transformers (Hugging Face)
Supported annotation tools: Label-studio, BRAT, MAE
The annotations are first converted to IE, then loaded by Dataset (PyTorch) to create training instances.
The raw text for information extraction is loaded and converted into IE. Then a fine-tuned NER model makes prediction on the IEs and outputs IEs with entities. An RE model then inpupts the IEs after NER and outputs IEs with entities and relations.
