Document-Based Sentiment Analysis on Financial Texts

Public opinion plays an important role in economic activities like investors’ decisions, and stock price changes. This study uses Sentiment Analysis as the technique of Natural Language Processing to identify economic texts’ sentiment. Sentiment Analysis is performed at the document level for each text. Term frequency and inverse document frequency (TF-IDF) algorithms are used for text summarization and document vectorization. Random Forest, SVM, and MLP algorithms are used to test the predictive performance of the model through accuracy, precision, recall, and F1-score. In the final stage, the Voting Ensemble of individual algorithms is combined using the majority vote rule for classification. The best-performing model is the Voting Ensemble with an overall accuracy of 71%. The need for sentiment analysis in economic activities will continue in the future as data generated from different sources rapidly grows.