Sentiment Analysis of Animated Film “JUMBO” on Twitter Using Random Forest and Semi-Supervised Learning

Abstract

This study investigates public sentiment toward the Indonesian animated film "JUMBO" using Twitter data and a semi-supervised machine learning approach. Two thousand fifty tweets were collected and preprocessed to remove noise, standardize text, and extract meaningful features. Data was collected between April 6, 2025, and May 13, 2025, following the film's official release on March 31, 2025, coinciding with its peak public discussion window. A semi-supervised learning strategy was applied, where 532 tweets were manually labelled into positive, neutral, or negative sentiment categories, mitigating the extensive need for labelled data. To address the class imbalance, the Synthetic Minority Over-sampling Technique (SMOTE) was employed. The labelled data were then used to train a Random Forest classifier, achieving an accuracy of 90% and balanced F1 scores across all classes. The model was subsequently applied to classify the remaining unlabeled tweets, which revealed a dominant proportion of positive sentiments toward the film. These results obtain strong public approval of "JUMBO" and demonstrate the effectiveness of combining machine learning with semi-supervised techniques for sentiment analysis, particularly in the context of local cultural products. This research can be an initial stage in a broader roadmap for analyzing the success factors of Indonesian animated films through AI-driven approaches.