Project for COMP-4740

Model created: train-model.py

python3 train-model.py

• make sure to download all the libraries listed

• lemmatization
• lowercase all words
• stopword and special character removal
• tokenized words and converted to sequences of indices

• CNN + MultiAttention Layer + FastText word embeddings
• Added hybrid pooling - Average and Max Pooling
• L2 regularization + Dropouts between layers
• Applied learning rate scheduler and early stopping for preventing excessive overfitting

• Accuracy
• Precison
• Recall
• F1 Score

Analysis of Deep Learning Model for Disaster Tweet Classification Speed Optimization Paper