Perceptron Sentiment Classification of Text

1. Single Perceptron Sentiment Classification

• Pre-processing: lowercasing, frequency of word count to limit vocabulary, NLTK Snowball stemming
• Unigram Feature
• Unigram and Bigram Features

2. Multilayer Perceptron Sentiment Classification

• two layer with tanh activation layer, with sigmoid activation for final output (used BCE loss)

3. Recurrent Neural Networks

• Bi-directional LSTM with "max pooling" activation to a linear fully connected layer. Then, this is fed to final Sigmoid activation.

4. RCNN

• Bi-directional LSTM with convolution and max pooling.