Sentiment Analysis of YouTube Comments Using Machine Learning Models

Abstract

The documentary video “115. You Are Human Too” from the #MenjadiManusia YouTube channel raises mental health issues with an empathic narrative approach. Social media plays a role in shaping public understanding, but opinions vary from support to stigma. This study analyzed the sentiment of 1,350 comments on the video using the YouTube API. Comments were classified into positive, negative and neutral sentiments using the IndoBERT model after preprocessing. Four machine learning algorithms were compared: Naïve Bayes, Random Forest, Support Vector Machine (SVM), and Extra Trees. Results showed that SVM had the highest accuracy (79.67%), followed by Random Forest (78.02%), Extra Trees (75.27%), and Naïve Bayes (70.33%). This analysis reveals patterns of public opinion on mental health, which can serve as a reference for academics, health practitioners, and policy makers in designing more effective communication strategies. In addition, this research is expected to increase public understanding of mental health and encourage more inclusive discussions on social media.