TextSentimentClassification

TextSentimentClassification, using tensorflow. Original Data

Data Preprocessing

Remove the letter whose number of repetitions is over 3 from a word...

Word Vectors Training

Using word2vec and GloVe to generate word vectors...

Models

Performance

TextCNN

Reference

Convolutional Neural Networks for Sentence Classification

A Sensitivity Analysis of (and Practitioners' Guide to) Convolutional Neural Networks for Sentence Classification

Model Architecture

Total 4 ways:

• CNN-rand
• CNN-static
• CNN-nonstatic
• CNN-multichannel

Performance

Choosing to use word vectors in a nonstatic way.

TextRNN

Model Architecture

Using bidirectional RNN, and then concatenating the output of the forward process and the output of the backward process...

CRNN

Reference

A C-LSTM Neural Network for Text Classification

Model Architecture

Using CNN to extract sentences with higher-level phrase representations, and then learning long short-term dependency with bi-RNN...

RCNN

Reference

Recurrent Convolutional Neural Networks for Text Classification

Model Architecture

In addition to implementing the same structure as the paper, using bi-LSTM or bi-GRU and then concatenating their outputs... RNN for capturing contextual information and max pooling used for judging which words play key roles in the task...

HAN

Reference

Hierarchical Attention Networks for Document Classification

Model Architecture

Transforming a sentence into a document consisting of sentences...

Ensembles

Bagging

Uniform blending...

Stacking

Using Logistic Regression as the level-2 classifier...

Performance