Standup 1

This is the first standup for our University of Michigan Master of Applied Data Science Capstone project. Dislike counts on Youtube videos are a useful signal for separating high quality videos from low quality videos or even potential scams. Youtube has recently removed the ability to see dislike count data publicly, and have disabled dislike count data in their API. Thus, our project seeks to understand the trends that influence dislike activity through analytical research and use those insights to generate a machine learning model to predict dislike counts or ratios to help alert users to potentially problematic videos based on the available data and features.

Standup 2

This is the second standup for our University of Michigan Master of Applied Data Science Capstone project. In this standup we discuss our progress since our last standup including how we overcome some obstacles, how we have begun analyzing the data, and an initial demo of a web app. Dislike counts on Youtube videos are a useful signal for separating high quality videos from low quality videos or even potential scams. Youtube has recently removed the ability to see dislike count data publicly, and have disabled dislike count data in their API. Thus, our project seeks to understand the trends that influence dislike activity through analytical research and use those insights to generate a machine learning model to predict dislike counts or ratios to help alert users to potentially problematic videos based on the available data and features.

Standup 3

This is the third standup for our University of Michigan Master of Applied Data Science Capstone project. In this standup we discuss our progress since our last standup including how we have processed our data to make it suitable for model training, initial testing of different models with the intention of picking a final model to use in production, as well as more progress on our web app.

We are web-scraping video comments which will be processed using sentiment analysis and used as extra features and joined based on video_id.
In the final pipeline, we will convert these experimentation notebooks to streamlined regular python files to aid in our overall pipeline.
We are using F1 score due to imbalanced data in our dataset. There is more liked/positive videos than negative ones.

Dislike counts on Youtube videos are a useful signal for separating high quality videos from low quality videos or even potential scams. Youtube has recently removed the ability to see dislike count data publicly, and have disabled dislike count data in their API. Thus, our project seeks to understand the trends that influence dislike activity through analytical research and use those insights to generate a machine learning model to predict dislike counts or ratios to help alert users to potentially problematic videos based on the available data and features.