Sentiment and emotion in Malay news: A comprehensive analysis using sentiment analysis

Abstract

Background and Purpose: Emotional framing of news can shape public perception and behaviour. This study examines sentiment in Malay-language headlines from Berita Harian (April 2021–April 2023) to reveal underlying emotions, recurring themes, and societal implications. Methodology: This research collected headlines from the Berita Harian archive, then applied tokenization, stop-word removal, and normalization. A pre-trained Malay sentiment transformer assigned initial labels (positive, negative, neutral), and a manually verified subset was used to train a Support Vector Machine (SVM). Model performance was measured on a test set via accuracy, precision, recall, and F1-score. Word clouds and count plots highlighted frequent sentiment features. Findings: The SVM achieved high precision and recall for positive sentiment (0.87/0.85) but lower recall for neutral (0.62), indicating challenges in neutral detection. Dominant topics included COVID‑19, PRU15, and mangsa. Contributions: By applying transformer labeling with SVM classification, this work extends sentiment analysis to Malay news media. It informs journalists and policymakers about emotional framing in Malaysian headlines. Keywords: Sentiment analysis, news headlines, TF-IDF features, Support Vector Machine (SVM), Malay language.