Classification of Television Programs Based on Public Opinion in Social Media Using Random Forest and Decision Tree

Abstract

One of the electronic media that provides a source of entertainment and information for its audience is television programs. The quality of television programs is assessed based on ratings. In addition, public opinion also has a crucial role, and this is because this opinion can be used as data to carry out sentiment analysis by classifying whether the opinion is included in the category of positive or negative sentiment. The problem requires an efficient decision-making system, such as a machine learning algorithm. This study aims to develop a classification of sentiment analysis using Random Forest and Decision Tree algorithms. The dataset used is a public opinion tweet dataset on Twitter for four television programs with five attributes: the number of retweets, id, sentiment, tv show, and tweet text. The results of testing the two algorithm models show that the Random Forest and Decision Tree models have different performance levels. The Decision Tree has lower performance with an accuracy of 74.66%, precision of 76.13%, recall of 78.75%, and an f1-score of 76.73%. Meanwhile, Random Forest has higher performance with an accuracy of 84.06%, precision of 83.65%, recall of 82.80%, and f1-score of 82.32% with a comparison of test data and training data of 20:80. Based on these results, the Random Forest model has the best performance compared to the Decision Tree in classifying television program sentiment based on user opinions in social media.