Why our named entity extraction API ?
Komprehend NER achieves State-of-the-art results on CoNLL 2003 test dataset with Precision 0.9, Recall 0.92 and F1-Score of 0.90. It uses character as well as word level embeddings and therefore, does not reply on POS labels to detect entities making it very useful to detect entities in user-generated content (Try “obama was the third president of america” in Komprehend and Spacy)
Komprehend NER does not lookup dictionaries like Freebase or DBPedia to identify the type of entities and therefore, is very fast to meet demands from various industries.
Komprehend NER can be customized with very few training examples and therefore, it can be adapted to any domain dataset.