Using Sentiment Analysis to Visualize Character Arcs in Avatar: The Last Airbender

End-to-end data science project in Python: 1) data scraping, 2) sentiment analysis, 3) and data visualization.

Motivation and project description

I was curious to see if sentiment scores could be used to visualize character arcs. If so, could sentiment analysis help writers evaluate character development in their work?

For this project, I analyzed one of my favorite TV shows Avatar: The Last Airbender. I used Jupyter Notebook for documentation. Follow along to see how to:

1. Scrape the web for episode transcripts with Beautiful Soup
2. Manipulate data with pandas and analyze character dialogue using VADER
3. Create interactive visualizations of the sentiment scores with Plotly Express

See the accompanying Medium blog post for detailed project tutorial.

Interactive visualizations

Without prior knowledge of the series, I could've guessed Azula is the villain by looking at the “Sentiments per episode” plot. Her trend line increases before a sharp decline towards the finale (typical for stories where the good guys win). Although sentiment progression is not a perfect proxy for character development, in the future it might be part of a larger algorithm that’ll help writers evaluate their work. See Medium blog post for further discussion.

Running the code

You must have Jupyter Notebook installed on your computer. Download atla_sentiment_analysis.ipynb to current directory and open Jupyter Notebook by running jupyter notebook in the command line.