Deep question-answering based on speech style in Avatar: The Last Airbender
Goal
The aim of this project was to create question-answering based on deep learning that answers in style of language used in Avatar: The Legend of Aang (officially known as Avatar: The Last Airbender).
Data
To solve the problem we used data available in Kaggle datasets, under the name Avatar: The Last Airbender (link)
Reproducibility and Quality Assurance
To reproduce results you have to clone the repository and run indicated source files.
Quality is assured by a CI/CD pipeline consisting of the following stages:
- linting by
flake8 - type checking by
mypy - testing using
pytest
Exploratory Data Analysis
Three of us, namely Ania, Konrad and
Witek are familiar with the cartoon, but one person, Jacek, has never watched even a single episode. Therefore, to
equalize expertise and to introduce you to the ATLA world we prepared EDA. In the jupyter notebook:
notebooks/exploratory_data_analysis.ipynb you can find our discoveries, among others, most frequent lemmas and
similarities between characters.