The above model works on a sentiment analysis on 1.4M tweets into 2 classes namely Posititve and Negative.

The embedding can also be downloaded from the same link.

tensorflow(2.1)

Download the data and embedding and put them in the manner shown below.

You can download the training dataset from the google drive link given here. https://drive.google.com/drive/folders/1JbaMolCoqdWw3hI2Yy0Keu5MLjsK6jyv?usp=sharing

Arrange your directory in the given manner.

Next run train.py to train the model which will generate the output_data folder with the tensorboard_logs output_logs and also the pickle tokenizer for inference.The saved_model is also saved in the folder.

The model architecture can be seen above.

To predict from a saved model use the predict.py function and to generate the predictions.

The model achieves around 80% accuracy with the given architecture and embedding. Increasing the size of the embedding may help generate better results. More embedding files can be dowloaded from the link https://nlp.stanford.edu/projects/glove/

You can change the embedding file or train your own embedding.That is something to be experimented with.