The dynamics of the financial market are heavily influenced by public perception reflected in economic news. Negative sentiment in news is often an early signal of market volatility. However, the high dimensionality and semantic ambiguity of financial text data pose challenges for automatic classification. This research implements a hybrid method of Latent Semantic Analysis (LSA) and Machine Learning for economic news sentiment classification. Using the Financial PhraseBank dataset, the text is processed through pre-processing and TF-IDF feature extraction before undergoing dimensionality reduction using LSA with 300 latent component via Singular Value Decomposition. The experimental result demonstrate that the Support Vector Machine (SVM) algorithm with an RBF kernel provides the best performance with an accuracy of 88.2% and an F1-Score of 85.8%. These findings prove that the integration of latent space in LSA effectively captures the semantic context of economic news, allowing it to be used as a reliable instrument for early market risk mitigation.
Dito et al. (Thu,) studied this question.