Sentiment Analysis of Indonesia’s Free School Lunch Policy Using LSTM and Word2Vec on YouTube Comments

Abstract

This study analyzes public sentiment toward Indonesia’s free school lunch policy using sentiment classification on YouTube comments. Data were collected from 5,640 videos, resulting in 485,097 comments, with 392,576 comments used for training and testing. The dataset was preprocessed through cleaning, tokenization, normalization, stopword removal, and stemming. Word2Vec was used for word embedding, and sentiment classification was performed using an LSTM neural network. The model achieved 82.56% accuracy on training data but 57.00% on manually labeled test data. The final sentiment distribution shows that negative sentiment slightly dominates, reflecting public skepticism about budget use and program effectiveness. Frequent keywords such as Indonesia, Prabowo, school, and corruption highlight key concerns. These results provide valuable insights for policymakers to improve communication and address public concerns. Future research should expand data sources, refine labeling, and test hybrid deep learning models to enhance classification performance.