To give content creators, both established and prospective, tools to study what kinds of content work
What it does
It takes any YouTube channel with a legacy username as input and calculates, based on their fifty most recent uploads, the average number of views per video, the average percentage improvement in viewcount due to text in the thumbnails, the correlation constant between the duration of the videos and viewcounts, and the correlation constant between the length of the titles and viewcounts. These data may be used by content creators to understand how they may optimise the way they create and present their content.
How we built it
We used Google's YouTube Data v3 API to extract the necessary information from YouTube channels, Google's Cloud Vision API to check the thumbnails of the videos for text, and a Pearson's Correlation Coefficient algorithm to calculate the relevant coefficients. We built the backend in node.js and the frontend in React.
Challenges we ran into
Authenticating the Google Cloud API and creating an express backend for the React front end were problematic because the APIs don't work with CORS.
Accomplishments that we're proud of
What we learned
What's next for Data Analytics for Content Creators
We look to expand how much information we draw from YouTube - in directions such as whether having a person's face in the thumbnail affects the viewcount. To make this a more complete tool, we also looked to implement functionality that used Twitter to ape what makes the best influencers so good. We only got as far as scraping sets of tweets from an account, and the next step would be training a text autocompletion model on this data.