CS291K

Sentiment Analysis of Twitter data using a combined CNN-LSTM Neural Network model

• Paper: https://www.academia.edu/35947062/Twitter_Sentiment_Analysis_using_combined_LSTM-CNN_Models
• Blog Post: http://konukoii.com/blog/2018/02/19/twitter-sentiment-analysis-using-combined-lstm-cnn-models/

Motivation

This project seeks to extend the work we did previously on sentiment analysis using simple Feed-Foward Neural Networks (Found here: paper & repo). Instead, we wish to experiment with building a combined CNN-LSTM Neural Net model using Tensorflow to perform sentiment analysis on Twitter data.

Dependencies

sudo -H pip install -r requirements.txt

Run the Code

• On train.py change the variable MODEL_TO_RUN = {0 or 1}
  • 0 = CNN-LSTM
  • 1 = LSTM-CNN
• Feel free to change other variables (batch_size, filter_size, etc...)
• Run python train.py (or, with proper permissions, ./train.py

Code Structure

• lstm_cnn.py : Contains the LSTM_CNN Model class to be instantiated.
• cnn_lstm.py : Contains the CNN_LSTM Model class to be instantiated.
• train.py : Main runner for the code. It instantiates a model, trains it and validates it.
• batchgen.py : Contains a couple of functions needed to pre-process and tokenize the dataset.