This repository contains the code which was used for the paper Sentiment Analysis of E-Commerce Reviews with Convolutional Neural Network using Natural Language Processing by Rohan Sawant. It started out as fork of Alexander Rakhlin's Convolutional Neural Networks for Sentence Classification in Keras .
The dateset used for in this repo could be found on Kaggle.
To train a new model set ... in sentiment_cnn.py
model_load = False
grid_search = False
To run the model on test set ... in sentiment_cnn.py
model_load = True
grid_search = False
To run grid search update the grid params in sentiment_cnn.py and set
model_load = False
grid_search = True
also
batch_size = [10]
epochs = [10, 15, 20, 25, 30, 35]
optimizer = ['Adamax']
The model CNN-Adamax-Final was trained with 25 Epochs, 10 Batch Size, Adamax Optimizer and 4 Convolutional Layers. This is where the paper diverged from the material which inspired it. The deeper neural network allowed it to better fit to the data.