Intent Classification with BERT

Objective

The goal of this project is to train and fine-tune a pre-trained BERT model on a text classification task.

Data

ATIS (Airline Travel Information Systems) Dataset: https://github.com/howl-anderson/ATIS_dataset

Data	Size
Training	4480
Validation	498 (10% of the original training set)
Testing	888
No. of classes	22

Model

Pretrained bert-base-uncased: https://huggingface.co/google-bert/bert-base-uncased

Results

Achieved a weighted f1 score ~ 97.7%

Device

1 NVIDIA GeForce GTX 1080 Ti

Software and Libraries

• Python 3.9
• Scikit-learn
• Tensorflow
• Pytorch
• Jupyter Notebook