Clustering Emotional Features using Machine Learning in Public Opinion during the 2019 Presidential Candidate Debates in Indonesia
Agus Sasmito Aribowo, Yuli Fauziah, Halizah Basiron, Nanna Suryana, Siti Khomsah · 2020
This research has produced a description of the emotions of streaming-video viewers of presidential candidate debates broadcasted on Youtube. In the first presidential candidate debate, the emotions of viewers were still neutral and tended to be feelings of pleasure and happiness. In the second to fifth presidential candidate debates, the dominant emotions were happy, angry, and sad. This research is known as emotion analysis, using comments from viewers of presidential candidate debates on Youtube as the data. Those comments were downloaded and pre-processed for data cleaning, emotion feature extraction, and clustering using K-Means based on six basic types of emotions: anger, sadness, happiness, fear, surprise, and disgust. The aims to be achieved are to determine a more homogeneous cluster for each opinion in the presidential candidate debate videos and to provide an emotional label for each cluster formed. The results of the research are five clusters that have distinctive homogeneity, namely happiness, anger, neutral, surprise-angry-disgust, and sadness. Each cluster member was labeled according to its characteristics. After being divided for each stage of the presidential candidate debate, it can be seen that the journey from the first debate to the next debate period tended to increase the emotion of anger and reduce the emotion of neutral.