Sentimen Analisis Berbasis Aspek terhadap Aplikasi Tiktok Berdasarkan Ulasan pada Google Play Menggunakan Metode Extreme Gradient Boosting (XGBOOST)

Abstract

Tiktok is a popular application that is in great demand in the world, especially in Indonesia. Tiktok allows its users to make videos that are up to 10 minutes long and can be accompanied by music, filters, and many interesting features that encourage the creativity of its users to become content creators. This TikTok application was officially launched in 2016 by Zhang Yiminy from China. This application can create long videos and can add many features, such as adding music to videos, changing voice, filtering, and adding effects and stickers. This TikTok application also provides coins that are used when other users use the live feature and support some of its users to shop on the TikTok feature. TikTok operates under the condition that it guarantees that it will not display negative content such as pornography and SARA. Unfortunately, there are no clear rules regarding the regulation and supervision of social media applications. Because of this, negatively charged content still often escapes scrutiny. Currently, there are many different opinions about the Tiktok application. Therefore the author views the problem as focusing on the need for more detailed information regarding aspects of the opinions expressed by the community, to see benchmarks for the community to increase their awareness of using the Tiktok application and as a benchmark for companies to carry out developments based on aspects that have been being delivered. Therefore we need an approach that can analyze the sentiment aspects of the Tiktok application based on reviews on the Google Play store. In this study, the system analyzes Tiktok users through the Extreme Gradient Boosting (XGBoost) method, in data processing using 6460 Data, using 3 Sentiments, namely Positive with a value of 1, Neutral with a value of 0, Negative with a value of -1. Where the comparison of Data Training and Testing is 8: 2, the process stages carried out in this research are Cleaning, Normalization, Case Folding, Stopword Removal, Stemming, tokenization, and TF-IDF. And get an accuracy of 88%.