We develop a deep convolutional neural network to perform sentiment analysis on movie reviews and classify them as negative/neutral/positive. We achieved a test accuracy of 70.658684%.
View our final report here: https://baranusluel.com/sentiment-cnn/
Our code is contained in the Juypter notebook project.ipynb. We have included
a saved model Sentiment_CNN_V1.h5 which yielded our best test results.
The notebook was executed on Google Colab, but it can be run locally as well. The path variables will need to be adjusted accordingly.
This project uses Keras with a Tensorflow backend. It also depends on numpy and scikit-learn.
The pretrained word2vec model (GoogleNews-vectors-negative300.bin.gz) will need to be downloaded.
It can be found here.
The dataset of movie reviews called scale dataset v1.0 should also be downloaded
from here and unzipped.