Hello Aakriti! I have observed the class distribution of each output category (you can find the class distribution graph in here: https://github.com/DanielDaCosta/RNN-Keras/blob/master/ML-Pipeline-RNN.ipynbs ), and then analyzed the proportion of each class. That's how I came up with the following weights:

weight_binary = {0: 0.5, 1: 7}

weight_categorical = {0: 1.4, 1: 0.43, 2: 7}