A Machine Learning Based Detection of Fake Engagement in YouTube Videos

Abstract

The availability of online video sites is exposed to huge amounts of user interactions, and as a result, spam comments and fraudulent engagements are likely to contaminate these platforms thereby subverting the popularity of content and altering the recommendation engines. Such malicious activities have been a significant problem in social media analytics, machine learning, and natural language processing to detect. The currently used solutions tend to use the common and traditional classifiers, like the Naive Bayes, Logistic Regression, and Support Vectors machines, which might not help in modeling the complex textual patterns within the massive comment data. This paper presents a machine learning-based system of identifying fake interest in YouTube videos through spam comment detection. The suggested solution will combine the preprocessing methods of the text with Term Frequency–Inverse Document Frequency (TF–IDF) feature extraction mechanism that is capable of turning the textual remarks into the numbers. A classifier based on Extreme Gradient Boosting (XGBoost) was then used to distinguish between spam and legitimate comments in the data. The dataset of YouTube spam comments with 1956 comments was evaluated experimentally. The accuracy of the proposed model was found to be 91.3 with a precision value of 0.96, a recall of 0.86 and an F1-score of 0.90 and the ROC analysis provided a value of 0.96 which is regarded as good classification. These findings indicate that text analysis processes supported by machine learning will be able to detect the presence of spam comments and minimize the presence of fake interactions, which will bring greater reliability and credibility to online video websites.