Toxicity Analysis and Sentiment Classification of Wonderland Indonesia by Alffy Rev using Support Vector Machine

Abstract

The music industry's increasing reliance on digital platforms like YouTube for dissemination raises concerns about the potential impact of music videos on viewer sentiment and well-being. This study seeks to assess the toxicity and sentiment of the Wonderland Indonesia music video by Alffy Rev through Support Vector Machine analysis, contributing to our understanding of the effects of music content on online audiences. This research addresses the challenge of sentiment classification in digital content by leveraging the Cross-Industry Standard Process for Data Mining (CRISP-DM) framework. The study aims to enhance sentiment classification accuracy by applying a Support Vector Machine (SVM) with a Synthetic Minority Over-sampling Technique (SMOTE) to address class imbalance issues. The research problem revolves around the need for robust sentiment analysis models capable of accurately discerning sentiment polarity within diverse datasets. Through the systematic application of CRISP-DM phases - business understanding, data understanding, data preparation, modeling, evaluation, and deployment - the study examines the efficacy of SVM with SMOTE in sentiment classification tasks. The findings demonstrate notable performance metrics, including accuracy (96.50%), precision (95.75%), recall (99.00%), and F-measure (97.34%). The AUC value substantially increases from 0.642 without SMOTE to 0.997 with SMOTE, highlighting its effectiveness in improving sentiment classification accuracy. In addition, The comparative analysis of toxicity values between the first and second videos demonstrates distinct patterns: the first video showcases a Toxicity score of 0.05290, with notable metrics such as Profanity registering at 0.04815. Conversely, the second video exhibits a slightly lower Toxicity score of 0.04744, with varying metrics such as Severe Toxicity at 0.01386.