Twitter Sentiment Analysis

Overview

This package will perform sentiment analysis on tweets or similar short texts. Pre-trained word embeddings from GloVe are used as a frozen input to Keras, afterwhich a CNN learns and predicts on the classification.

Resulting accuracy: *~ 79%*

Usage

1. Download the GloVe embedings for twitter, unzip into a /glove directory.

    wget http://nlp.stanford.edu/data/glove.twitter.27B.zip
   

2. Download the twitter sentiment data set into / directory.

    wget http://thinknook.com/wp-content/uploads/2012/09/Sentiment-Analysis-Dataset.zip
   

3. Run [Optional since model.h5 and weights.h5 have been provided]

    python train.py
   

   Which will create a model.h5 and weights.h5 files.

4. Run

    echo "This is a sample tweet to predict on" | python predict.py
   

   Or

    cat file-containing-one-tweet-per-line.txt | python predict.py
   

Reference

• Using pre-trained word embeddings in a Keras model
• Twitter sentiment data set
• GloVe embeddings