NeuroMovieSentiment

Teammembers

• Daniela Ruchser
• Daniel Betsche
• Alan Mazankiewicz

Repository Overview

Experiment introduction

Installation

Running the Experiments

References

Experiment

The conducted experiment within this repository is to classify sentiment on Movie Reviews extracted from rotten Tomatoes. The data is provided by this Kaggle Challenge

The task is build a neural network-based model to predict the reviews as of five sentiment categories:

• 0: Negative
• 1: Somewhat Negative
• 2: Neutral
• 3: Somewhat Positive
• 4: Positive

There are several Experiments nested within the modelle folder. We build 3 models from which we consider one (CNN in modelle/main_model) as our main solution while we use the other two (two LSTM variants in modelle/side_model) to compare against. Further, we compare against the model described in Socher et al., 2013 (Stanford CoreNLP in modelle/external_model). Beside this we did some exploratory analysis of the data in analysis.

Installation

It is recommended to use a Kaggle-Kernel / Kaggle script for the provided experiments. In case you still want to run these on your local machine continue reading

After you clone the repository it is advised to create a virtual environment and install the required packages within

Using Virtualenv:

It is assumed you have Virtualenv alread installed. If not please follow these instructions

mkdir movie_sentiment
virtualenv movie_sentiment
source movie_sentiment/bin/activate
pip install -r requirements.txt

Using Conda

It is assumed you have Conda alread installed. If not please follow these instructions

conda create --name movie_sentiment python=3.7
source activate movie_sentiment
conda install --yes --file requirements.txt

Running

• To run the experiments on Kaggle-Kernel upload the given notebook / script to Kernel, activate GPU and internet connection and run.
• To run on local machines follow the instructions in Installation and run scripts using the CommandLine and notebooks using Jupyter Notebook.

Note: In case the Standord CoreNLP should be run in addition to Installation download 'stanford-corenlp-full-2018-10-05', (see https://stanfordnlp.github.io/CoreNLP/index.html#download) (only tested on local machines)

References

Socher, R., Perelygin, A., Wu, J., Chuang, J., Manning, C.D., Ng, A. and Potts, C., 2013. Recursive Deep Models for Semantic Compositionality over a Sentiment Treebank. In Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing (pp. 1631-1642).

Kim, Y. (2014). Convolutional neural networks for sentence classification. arXiv preprint arXiv:1408.5882.

https://github.com/bentrevett/pytorch-sentiment-analysis/blob/master/4%20-%20Convolutional%20Sentiment%20Analysis.ipynb

https://github.com/bentrevett/pytorch-sentiment-analysis/blob/master/2%20-%20Upgraded%20Sentiment%20Analysis.ipynb