Sentiment analysis experiments

This repository contains code for performing experiments checking quality of sentiment analysis of different ML algorithms and datasets.

Used datasets

Setup

To test notebooks, you need these datasets unpacked and saved to 'data' directory:

Amazon Fine Food Reviews IMDB

In notebooks, I assume following files in data directory:

$ cd data
$ ls
fine_foods_reviews.sqlite  # renamed from database.sqlite
aclImdb                    # extracted IMDB dataset

This project is designed to be run using docker. CPU version is much easier to run:

# we'll creating symlink to CPU docker-compose to be used by default
ln -s docker-compose.yml docker-compose.cpu.yml
docker-compose up

# jupyter notebook should be listening on localhost:8888, with all examples ready to be run.

GPU version requires nvidia-docker and nvidia-docker-compose. After installing, run:

# we'll creating symlink to GPU docker-compose to be used by default
ln -s docker-compose.yml docker-compose.gpu.yml

docker-compose build

# run just one time to ensure proper volumes are created
nvidia-docker run --rm nvidia/cuda:8.0-cudnn6-devel-ubuntu16.04 echo "It's working"

# start using nvidia-docker-compose
nvidia-docker-compose up

After opening localhost:8888, there should be a file browser allowing to run Jupyter Notebooks.

Repository structure

Notebooks used for testing:

IMDB Reviews models generation and results: src/text_sentiment_analysis_food.ipynb
Amazon Fine Food Reviews models generation and results: src/text_sentiment_analysis_imdb.ipynb
Results comparision: src/results_comparision.ipynb

Web application code is in src/app.py.

There is a quite large code base shared between notebooks and application, available inside src/shared directory. Model definitions can be found in src/shared/models.py.

Valian/nlp-sentiment-analysis

Sentiment analysis experiments

Used datasets

Setup

Repository structure