Keras Models for ATIS Dataset

Spoken Language Understanding(SLU)/Slot Filling in Keras.

There are 4 models created in Keras to solve the Airline Travel Information System(ATIS) dataset.

1. Simple RNN
2. LSTM
3. CNN with GRUs to Capture Future Contexts
4. BiDirectional LSTM

Here is an example sentence and its labels from the dataset:

Show	flights	from	Boston	to	New	York	today
O	O	O	B-dept	O	B-arr	I-arr	B-date

Results

Model	RNN	LSTM	GRU-CNN	BD-LSTM	BD-CNN*
F1 Score	93.02	94.24	94.72	95.43	95.61

*Model Results are from Sequential Convolutional Neural Networks for Slot Filling in Spoken Language Understanding, Ngoc Thang. Vu. and hasn't been evaluated in the code.

GRU CNN model, training and evaluation script forked from https://github.com/chsasank/ATIS.keras