Comparing Semi-Supervised Machine Learning Algorithms to Sentiment Analysis on English and Filipino Tweets during the POGO Probe

Semi-Supervised Machine Learning was created as a hybrid approach where training data are both labeled (supervised) and unlabeled (unsupervised). This research compares three algorithms that make use of the semi-supervised machine learning approach on both English and Filipino tweets. This study involves tweets collated during the height of the Philippine Offshore Gaming Operations investigation involving (former) Mayor Alice Leal Guo of the Municipality of Bamban in the Province of Tarlac. Through Natural Language Processing, both English and Tagalog tweets were annotated, processed, and trained to be classified into three polarities: positive, neutral, and negative. With 70% unlabeled data and comparing Self-Training (with Multinomial Naïve Bayes), Label Propagation, and Label Spreading, Label Propagation bested all others and results yielded an accuracy of 89.43% on testing which outdone previous studies on English and Filipino Twitter data.