/Sentiment-Analysis-with-Convolutional-Networks

Sentiment Analysis with Convolutional Networks

Here is one of my submissions to Kaggle challenge 'Bag of Words meets Bags of Popcorn'.

It is based on the idea of combining pre-trained word2vec embeddings with convolutional networks proposed by Yoon Kim [http://arxiv.org/abs/1408.5882].

The code consists of two IPython Notebooks:

  1. Process Kaggle Dataset Train+Test.ipynb contains data pre-processing.

  2. Train CNN IMDB.ipynb implements convolutional network with one convolutional layer.

This model (trained for 3 epochs) yields AUC = 0.96823 (on test data).

Ensemble of three convolutional networks (having different number of convolutional layers and feature maps) gives AUC = 0.97310.

Dependencies