FINANCIAL NEWS SENTIMENTS: A COMPUTATIONAL LINGUISTICS ANALYSIS

Abstract

Financial news is crucial for forecasting market trends and making informed investment decisions, but analyzing large volumes of such news is time-consuming and energy sapping. Computational linguistics, or specifically known as Natural Language Processing (NLP) addresses this by automatically detecting sentiments through either heuristic/lexicon-based approaches or machine learning techniques. This study developed a system that uses both machine learning and heuristic-based techniques to analyze the sentiment of financial news articles. The research utilized a dataset consisting of 4,846 records of financial headlines. The dataset was cleaned and preprocessed by removing duplicates, lemmatizing the words, and removing stop words. Then, four machine learning algorithms were experimented with, including logistic regression, linear SVC, random forest, and multilayer perception. The results of the experimentation showed that logistic regression performed best with an accuracy of 76.65% and an F1 score of 71.37% when utilizing the bag of words representation of the text and heuristic feature to learn. The conclusion of the research showed that computational linguistics has potential in analyzing the sentiments of financial news. It was recommended, among other things, that combining lexicon-based approaches and machine learning techniques improves sentiment detection accuracy from financial news compared to using either technique alone. Future researchers should consider this as part of their research.