<![CDATA[Stunting is a serious health issue in Indonesia, including in East Nusa Tenggara Province (NTT), which requires an analysis of public perception to support data-driven policies. This study proposes sentiment analysis using deep learning and machine learning approaches to classify public opinions regarding stunting from social media/online platforms. It aims to evaluate the performance of the BERT (Bidirectional Encoder Representations from Transformers) and SVM (Support Vector Machine) models in identifying sentiment (positive, negative, neutral), compare the advantages of BERT (transformer-based) and SVM (traditional machine learning) for sentiment classification tasks, and analyze the linguistic and contextual factors influencing sentiment polarity through text feature extraction. The research methods include collecting text data from digital platforms, text preprocessing, and model training with BERT embeddings as input features for SVM. The results are compared with traditional baselines (TF-IDF and word2vec) to measure accuracy improvement. The evaluation results show that for Negative Sentiment (86 tweets) Precision: 58%, Recall: 58%, F1-Score: 58%, Accuracy: 100%. Neutral Sentiment (814 tweets) Precision: 30%, Recall: 20%, F1-Score: 25%, Accuracy: 100%. Positive Sentiment (100 tweets) Precision: 60%, Recall: 75%, F1-Score: 68%, Accuracy: 100%. Meanwhile, SVM with various kernel types showed performance differences in sentiment classification.]]>
Copyrights © 2025
