This study examines how media outlets frame sentiment in news headlines and explores the main themes within media coverage using machine learning methods. A sentiment analysis model was trained and tested, achieving an average validation accuracy of 67.7% and a macro-average F1-score of 0.67 across negative, neutral, and positive sentiment classes. The model showed strong performance in identifying negative sentiment, while neutral and positive classes were harder to classify. Latent Dirichlet Allocation (LDA) topic modeling produced five dominant themes, which were mostly related to influencer culture, social media trends, and corporate marketing. In contrast, issues such as fake news, censorship, and press freedom were only sparsely represented, indicating a limited focus on media-related concerns in the dataset. Correlation analysis between sentiment and media issue coverage revealed a weak negative relationship, showing that these issues had little effect on overall sentiment patterns. Where they were present, the coverage of media issues tended to carry a negative tone. The results demonstrate the usefulness of machine learning in analyzing media sentiment and topic structures while pointing to the underrepresentation of critical media issues in headline coverage.