From Hype to Harm: Automated Detection of Clickbait and Misleading Videos on YouTube Using Hybrid NLP and Engagement Metadata

Abstract

YouTube is a source of information for people all around the world but it is still very easy for
people to post fake or misleading videos that trick people into watching them. These videos often
use language that's not true and they make people think they are more popular than they really are.
Most of the ways we have now to detect these videos either only look at the words used in the video
title and description, or they use big computer systems that take a lot of time and money to run.
This paper is talking about a way to detect these fake videos that is faster and cheaper. It uses a
combination of things like the words used in the video title and description and things like how
many people like and comment on the video and how many people watch it over time. We get this
information from the YouTube Data API v3.
We use computer models like Random Forest and Logistic regression to look at all this information
and figure out if a video is fake or not .We tested this way on a big group of 25,000 videos and it
worked really well. It was right 95.1 percent of the time. This is better than looking at the words
used in the video title and description or just looking at the other information like likes and
comments. It is also faster than the computer systems we use now.
The main thing we are trying to do is detect clickbait videos on YouTube and we are using a new
way that combines machine learning and natural language processing. We are also looking at how
to stop the spread of information, on YouTube and we think this new way can help with that.