Sentiment Analysis on Investment Education from Twitter using Ensemble Learning

People’s perspectives can be found on various platforms, including Twitter, Instagram, Facebook, and Youtube, as a result of the recent rise of social media as a significant means of education and communication. This research makes use of a dataset obtained from Twitter that contains 700 posts discussing various aspects of investment education. Natural language processing tasks, including text analysis, have benefited greatly from machine learning and deep learning approaches due to their adaptability, effectiveness, and promise. We adapted the knowledge of native and linguistic experts in Indonesian language to manually confirmed whether a tweet is Positive or Negative. Manual annotation and text/emotion analysis have both seen extensive application of machine learning techniques. One model of learning model SVM as base learner has also demonstrated great classification accuracy, and its different kernels have been utilized for sentiment categorization. To effectively create and test the learner, we used fastText and Multilingual Sentence BERT, the two methods of feature extraction. All of the SVM learners with different kinds of kernels have been tested in terms of key performance indicators such as Accuracy, Precision, Recall, and F1-score. In our experiment, SVM with Polynomial kernel acquired the greatest accuracy of 98.1%.