Team 20's application for the 2020 Teens In AI Global COVID Hackathon.

MoodSmart was developed in Python 3.7. We highly recommend creating a separate Conda environment using the YML file provided in the repo and the following command:

conda env create -f teensinai.yml

This will automatically install all the required modules. If you don't have a Conda distribution installed in your system, we recommend installing Miniconda following the instructions at this page: https://docs.conda.io/en/latest/miniconda.html.

Once the environment is available, activate it as follows:

conda activate teensinai

To run the training of the deep learning models we use for tweet sentiment analysis, you will need the following two data sources.

1. The Sentiment140 data set, containing 1.6 million tweets labelled as either positive or negative. It can be found at this page on Kaggle: https://www.kaggle.com/kazanova/sentiment140.

2. A set of pre-trained word embeddings from Google News, also available on Kaggle: https://www.kaggle.com/umbertogriffo/googles-trained-word2vec-model-in-python.

These files should be downloaded and placed in a subfolder called data. They are only required to train the models. They are not required to use the Flask API.

To start serving the trained tweet sentiment analysis model, open a Terminal, navigate to the backend subfolder and run:

flask run --without-threads

The option --without-threads is required due to a bug in the latest version of Keras. This will start the API at the localhost address 127.0.0.1, port 5000.

To query the MoodSmart Flask API, make a POST request to the following endpoint:

http://127.0.0.1:5000/predict

The tweet you want to classify should be sent to the API via JSON in the following format:

{'tweet': "MoodSmart is the coolest thing ever!"} 

Upon successful completion of your request, you will receive a JSON response like this:

{'prediction': 0.9921677708625793}

This value represents the probability that the tweet is positive according to the model. Therefore it will be a number between 0 and 1, where 0 indicates really negative and 1 indicates really positive.