Automatically detect in-video YouTube sponsorships, self/unpaid promotions, and interaction reminders using transcripts. Currently being used to find segments that must be reviewed by a human and made more precise.

NeuralBlock (NB) is a neural network built using Keras/Tensorflow that detects in-video YouTube sponsorships based off of captions. There is support for both predicting whether or not a text excerpt is a sponsorship (spot) or whether or not this word in the sequence is part of a sponsorship.

This works and you can test it out on https://ai.neuralblock.app/ or deploy it yourself.

For now, the plan is to use this as a moderation tool, however, it may also be used in the future to act as a tool to help with a human making submissions.

Right now, there are issues with YouTube rate-limiting the server when trying to pull a lot of captions.

Deep learning-based solution for identifying sponsored content segments on YouTube videos based off of video frames. This project was completed in Autumn 2020 for Stanford's CS 230 by Nikhil Athreya, Cem Gokmen and Jennie Yang.

It's not really deployable (quite intensive to do this over frames) but was an interesting research project

Detect sponsored segments in Youtube videos With the power of machine learning!

Seems to be an ML detection based on transcripts

https://phiresky.github.io/youtube-sponsorship-stats/

Underscore_: https://www.youtube.com/watch?v=ht_yTaAzdF0 (SponsorBlock dataset not used for final analysis, but discussed as one of the original ideas)

Sylvqin: https://www.youtube.com/watch?v=cjSSxMksHVE Uses SponsorBlock data along with other data for analyzing sponsors on French YouTube