Design of a Sentiment Analysis System for Indonesian-Language International Conflict News using Naïve Bayes and SVM

Abstract

The rapid growth of digital news content, particularly coverage related to international conflicts, has created an urgent need for sentiment classification systems capable of operating automatically and objectively. This study designs a sentiment analysis system for Indonesian-language international conflict news using two classification algorithms, namely Naïve Bayes and Support Vector Machine (SVM), while also comparing their effectiveness. A total of 339 news articles sourced from Reuters, BBC, CNN, and Al Jazeera were used as the dataset, consisting of 146 negative, 123 positive, and 70 neutral sentiment texts. The data processing stages included preprocessing (case folding, tokenizing, stopword removal, and stemming), TF-IDF feature weighting, and classification using an 80:20 train–test split scheme. The system was developed using Python and deployed through the Streamlit framework as a web-based interface. The experimental results indicate that SVM achieved an accuracy of 82.35%, with more balanced precision, recall, and F1-score values across all classes, outperforming Naïve Bayes, which achieved an accuracy of 77.94%. The main novelty of this study lies in the use of complete news articles from credible international journalistic sources as the object of analysis, unlike most previous studies that primarily relied on short-form social media data. The study concludes that SVM is more effective for sentiment classification of Indonesian-language international conflict news and recommends expanding the dataset and exploring semantic representation methods such as IndoBERT to improve performance in future research.