This project is for the Machine Learning Practical (WBAI060-05) course, as such we cannot accept outside contibutions at the moment.

NLP Wizards: Emoji Prediction

In this project, we will be predicting the emoji for a given tweet. Our current approach uses a neural network classifier which is trained on the word2vec sentence embeddings of the tweets.

Pre-requisites

To install all the dependencies, run the following command using pipenv:

pipenv install

Project Structure

We currently have 3 main files in the project:

• preprocess_dataset.py: This file is used to clean the raw data and store it in a CSV file.

• create_embedding.py: This file is used to generate the sentence embeddings for the classifier. The embeddings are stored as numpy files.

• create_model.py: This file is used to train the classifier with a grid search and store the model.

Running the project

To run the project, you can use the following commands:

pipenv shell
python preprocess_dataset.py
python create_embedding.py
python create_model.py

We will be consolidating all the commands into a single file in the future.

Starting the API

To start the API, you can use the following command:

# It can take a while to start the API, so please be patient.
# It loads the model and the embeddings into memory.
uvicorn api:app --reload

You can access the Swagger Documentation at http://127.0.0.1:8000/docs and the API at http://127.0.0.1:8000/get_emoji?text=....

Starting front-end

To start the front-end, you can use the following command:

streamlit run streamlit_demo.py

Authors

• Mansur Nurmukhambetov
• Jeremias Lino Ferrao
• Juriën Michèl Schut