Sentiment Analysis Terhadap Perspektif Warganet Atas Tragedi Kanjuruhan Malang di Twitter Menggunakan Naïve Bayes Classifier

Abstract

Media sosial Twitter adalah salah satu tempat bagi warganet dari seluruh dunia untuk menyampaikan perspektif mereka, sebuah insiden yang terjadi di Stadion Kanjuruhan Malang pada tanggal 01 Oktober 2022 sedang hangat diperbincangkan, sehingga memunculkan berbagai perspektif yang memicu timbulnya pro-kontra di masyarakat. Atas dasar itu untuk mengklasifikasikan perspektif positif atau negatif warganet di Twitter, maka dilakukan analisis sentimen menggunakan Naïve Bayes Classifier. Analisis sentimen dilakukan dengan mengambil tweet warganet di Twitter dengan hashtag UsutTuntasTragediKanjuruhan yang diambil 1.500 tweet untuk dijadikan dataset. Preprocessing data terdiri dari Annotation Removal, Remove Hashtag, Transformation Remove Url, Regexp, Indonesian Steaming, Indonesian Stopword Removal. Hasil analisis berjalan dengan baik dengan nilai akurasi 77,67%, kemudian nilai precision sebesar 77,19%, nilai recall sebesar 78,50%, dan nilai AUC 0.820 (good classification).

The social media site Twitter is a place where Internet users around the world can exchange perspectives on current discussions. One of them is football; this sport is a hobby that is loved by all corners of the world, including the people of Malang. With their love for this sport, they call themselves Aremania, namely Arema Malang team supporters, but a dark incident occurred. The Kanjuruhan at Malang Stadium on January 10, 2022, raised different views from all Twitter user accounts, which led to an increase in tweets and became a trending topic at that time. To develop different perspectives based on what brings advantages and disadvantages to the community, a procedure was applied to classify Twitter users' positive or negative perspectives through sentiment analysis with the Nave Bayes classifier. Sentiment analysis was carried out by indexing Twitter user tweets with the hashtag "UsutTuntasTragediKanjuruhan," crawling data from 1,500 existing tweets as a dataset, after which the data to be processed is identified. (labeling) for the next step, namely stage Data preprocessing includes annotation removal, hashtag removal, URL removal, regexp, Indonesian steaming, and Indonesian stopword removal, as well as operators' smote upsampling. Making a confusion matrix that shows the final result of the analysis is going well, namely the value accuracy of 77.67%, the value precision of 77.19%, the value recall of 78.50%, and the value AUC of 0.820 (good classification).