Publication

Abstract : YouTube is an open platform where anyone can create content and share it with the world via their own YouTube Channel. YouTube offers a wide range of content for its audience of all ages, including comedy shows, vlogging, cooking recipes, hacks, unboxing videos, and other content are available. If you’re a frequent YouTube viewer and have spent time reading YouTube comments, you might have noticed a repetitive pattern, along with abuse or childish insults and dubious viewpoints, which is commonly referred to as spam. The presence of such spam comments has a negative impact on a channel's reputation as well as the experience of normal users. YouTube has addressed the problem with very limited methods, such as YouTube Studio, which offers a “Report” or “Report as spam or abuse” option for comments, allowing the community to control the number of spam comments left on videos. While reviewing comments in YouTube Studio, creators can also report spam. However, these methods have attested to be not fruitful, as spam creators discovered methods to circumvent such heuristics. As a result, in this paper, we use six traditional machine learning algorithms to detect spam comments: Gaussian Naive Bayes, Support Vector Machine, Linear Classifier, AdaBoost, Random Forest, and Decision Tree Classifier. The data for the model was obtained from the Kaggle Repository, and the experiments were carried out using Google Collaboratory. Thus, we can achieve an accuracy of 91.39% with the AdaBoost Classifier, beating the current course of action by around 20%, and have shown to be extremely successful at detecting and removing spam comments.