YouTube Videos Clickbait Classification Utilizing Text Summarization and Similarity Score via LLM

Clickbait detection is used to check for clickbait, preventing users from wasting their precious time watching videos that might have misled them. Past researches have used methods such as Deep Learning algorithms that take into account the thumbnail of the video, the statistics, and the comment section. There are also researches that uses LLMs for detection. With this in mind, the author introduces a new way of clickbait detection through the combination of both YouTube statistics and LLMs, as well as the addition of Youtube transcripts as one of the determining factor. This is done with the use of text summarization and OpenAI's ChatGPT. The video transcript is extracted and summarized. ChatGPT will then create a new title suitable for the summarized transcript. The made up title is then compared to the original title and be given a score based on their similarity and used as a new feature for the model. ChatGPT will also be asked to directly predict the presence of clickbait directly from the title and the summarized transcript. All the features are used to create a new machine learning model. The algorithms used for the classification include Logistic Regression, Naïve Bayes, Random Forest, Multi Layer Perceptron, and Support Vector Machine. Random Forest achieved the highest f1-score out of all the models with the score of 87%.