There are two notebooks in this repository that use two different methods to predict YouTube dislikes. One of them uses VADER to do sentiment analysis on textual features, and the other one uses the bag-of-words technique. The bag-of-words (BoW) technique showed less mean absolute error.
Here is the Medium article I wrote about this project:
https://medium.com/@ramtiin/predicting-youtube-dislikes-via-catboost-and-sentiment-analysis-ec3252ce8317