Toxic_Comment_Classification

My Trial to tackle the Kaggle Toxic Comment Classification Competition

I built a model that calculates the probability of a comment belonging to any of the mentioned classes, I used XGBoost after generating feature vectors using GLove and Google news Word2Vec

I got a total AUC of 0.82

Resources needed:

• Download data from kaggle competition page here
• Download GLove Word Vectors here, choose the 300d.480B model
• Download GoogleNews Word Vectors here
• To use the Keras model built in the file example_to_clarify.py, you need to download the 20 Newsgroup dataset

Note:

final_try.py 

file is an implementation to XGBoost algorithm on the same data

To Do::

1. You definetly can make much more hyperparameter optimization epecially regarding the LSTM model. for example: You can try playing around with max_features, max_len, Droupout_rate,size of the Dense layer, etc...

2. You can try differnt feature engineering and normaization techniques for the text data

3. In general try playing around with parameters like batch_size, num_epochs and learning_rate

4. Try to use differnt optimization function, maybe Adagrad ,Adadelta or sgd