Ekstraksi Opini Negatif dan Klasifikasi Sentimen Pengguna Menggunakan Latent Dirichlet Allocation dan Support Vector Machine pada Program Danantara di Media Sosial X

Abstract

This study aims to analyze public sentiment toward the Danantara Program on social media X by combining Latent Dirichlet Allocation (LDA) for topic extraction and Support Vector Machine (SVM) for sentiment classification. The dataset consisted of tweets related to Danantara, which underwent several preprocessing stages, including case folding, cleaning, normalization, stopword removal, stemming, and tokenization. Topic modeling was performed exclusively on negative sentiment tweets using the LDA method, while sentiment classification focused on positive and negative classes using the SVM algorithm with an RBF kernel. The optimal number of topics was determined using the coherence score, where the highest value of 0.5798 was achieved with seven topics. The extracted topics revealed major public concerns regarding transparency and accountability of state-owned enterprises, natural resource management, investment risks, state budget allocation, strategic project management, stock market stability, and social impacts of government policies. For sentiment classification, the SVM model was evaluated using three train-test split scenarios (90:10, 80:20, and 70:30). The best performance was achieved with a 90:10 split, resulting in an accuracy of 79%, while cross-validation scores ranged from 70% to 72%, indicating good generalization ability without signs of overfitting or underfitting. The findings demonstrate that the integration of LDA and SVM is effective for identifying dominant issues and classifying public sentiment toward government programs on social media.

Keyword: Sentiment Analysis, Topic Modeling, Latent Dirichlet Allocation (LDA), Support Vector Machine (SVM), Danantara Program, Social Media X, Public Opinion.