Visualization Project: Global Analysis of Trending YouTube Data

Team Members

• Fereshteh Aghaei
• Jen Mahon
• Julia Squeri
• John Shuford

The final version of this project is deployed using AWS & a custom domain here.

Project Overview

Dataset

The data used for this project can be found here. The final dataset included the data for 10 counties (N= 45,698). Individual CSVs were downloaded, cleaned, and merged together for analysis. The original CSV files for each country can be found in the 'Data' folder, and the cleaned CSVs used for analysis can be found in the 'CSVs' folder. The Jupyter Notebooks used to clean and transform the data can be found in the 'Notebooks' folder.

Countries Analyzed

• US
• Russia
• Canada
• Great Britain
• France
• Germany
• Japan
• South Korea
• India
• Mexico

Specific Project Requirements

1. We built on a previous project where we used SQL/PGAdmin to extract, transform, and load data for US and Russia YouTube Statistics. We added eight more countries to our analysis. The schemas created to load our data into the database can be found in the 'SQL' folder.
2. We chose to create a custom, creative visualization project using JavaScript and D3.
3. We used three.js as our new JavaScript library to create a 3D representation of four of the trending YouTube videos.
4. Our final dataset included 45,698 records from a total of 10 countries.
5. Our final website include multiple forms of user driven interaction, such as a dropdown menu on the main page dashboard and on our bubble chart to navigate through and view the different data for each country. We incorporated multiple interactive visualizations where the user can select what countries data to view, or which statistics to see. Our website itself also has an interactive menu/navigation bar where the user can navigate through the website/different visualizations.
6. Our final project includes four unique visualizations, and one dashboard of different video metrics for each country. Our bubble chart visualization as well as our dashboard of country metrics include 10 different views, one for each country. Our bubble chart visualization also includes two different views for each country, one that includes every video, and one that seperates the videos into categories. Our interactive bar chart allows the user to navigate between three different metrics (average likes, dislikes, and views) and see the data displayed for each country.
7. Our final project also includes an 'About this project' page with a description of the project, dataset, and team members. This page also includes two word cloud visualizations created using Matplotlib that show some of the most common words used in the video titles in our dataset. The code for these word clouds can be found in the 'CategoryClassifier' folder in the Jupyter Notebook.

Machine Learning Model

For the final version of this project, we created a Naive Bayes model that allows a user to enter a hypothetical video title and see what category it would fall into. The model was trained on title and category data from the US, Canada, and Great Britain. After we fit, tested, and trained our model, we created a web form that would allow the model to take a title input and return a prediction of video category, such as 'News & Politics', 'Entertainment', etc. The Jupyter Notebook code for this model can be found inside the 'CategoryClassifier' folder. The python scripts, Flask application, and saved models for the back-end of our application can be found in the 'VisualizationSite_Final' folder inside the 'classifier' folder. As of now, the model will only work on our website if the user runs the 'app.py' script on their local machine. As a future direction, we would like to get the model/application working through AWS so any user can run the model without having to start it on their local machine.

Source Codes

For three of our visualizations, we adapted publically available code libraries for use with our own data/styling.

• The code used for creating our bubble chart can be found here.

• The code used for our map can be found here.

• The code used for our 3D Box can be found here.

• The final project can be found here.