Comparison of Classification Methods on Twitter Sentiment Analysis of PDAM Tugu Tirta Kota Malang

Abstract

The Regional Drinking Water Company (PDAM) Tugu Tirta is a public service company in Malang's drinking water distribution field. The company uses a customer complaint feature that is provided on the website. However, only a few people know about it and use it. From this problem, the researcher uses social media data, namely Twitter, to explore data sources and collect feedback tweets from the customer. However, analyzing the sentiment of the 1000 data used is elusive. The tweets contain unstructured text, so the researcher applies the labeling from the dataset, preprocesses the text, and then extracts the tweets by applying the classification methods by comparing Support Vector Machine (SVM), Naive Bayes (NB), Random Forest (RF), Logistic Regression (LR), Short-Term Long-Term Memory (LSTM), and Indonesian BERT to achieve highly accurate results. The tests with six methods show that Logistic Regression and Indonesian BERT are the best methods, with an accuracy of 85%. In this study, we obtained an effective algorithm to classify a comment as positive, negative, or neutral related to the Tugu Tirta Regional Drinking Water Company (PDAM).