abstract-dex/named-entity-recognition

Named Entity Recognition using HuggingFace Transformers

Named Entity Recognition

Named Entity Recognition (NER) is a subtask of information extraction that locates and classifies named entities mentioned in unstructured text into pre-defined categories such as the person names, organizations, locations, etc.

This task is done with the help of the Hugginface libraries: transformers and datasets

Dataset

The dataset used for this task is the CoNLL-2003 dataset which is available in the datasets library. The dataset is already split into training, validation, and test sets.

Data description

• id: a string feature.
• tokens: a list of string features.
• pos_tags: a list of classification labels (int).
• chunk_tags: a list of classification labels (int).
• ner_tags: a list of classification labels (int). Full tagset with indices:

Model

The model used to perform the NER task is the BertForTokenClassification model which is a BERT model pre-trained for token-level classification tasks. The model can be found here.

This model is then fine-tuned on the CoNLL-2003 dataset.

Training

The model is trained using the Trainer class from the transformers library. The training is done on the CoNLL-2003 dataset with the following hyperparameters:

• batch_size: 16

• learning_rate: 2e-5

• epochs: 3