Multi-aspect Sentiment Analysis of Tiktok Application Usage Using FasText Feature Expansion and CNN Method

Abstract

Among the many social media platforms that have emerged, TikTok is a platform that has the most significant number of subscribers compared to other platforms. However, not all reviews given by TikTok users are good reviews and reviews are often found with slang and not all reviews have real meaning, therefore sentiment analysis is needed for these problems. These reviews will later be analyzed for sentiment according to predetermined aspects, namely feature aspects, business aspects, and content aspects based on reviews written on the Google Play Store, using data crawling techniques and will pass the preprocessing and weighting stages. The weighting method used is Term Frequency-Inverse Document Frequency (TF-IDF). Then, the sentiment analysis process will use the Convolutional Neural Network (CNN) method, and feature expansion will be carried out to determine what words are interrelated with certain words. The purpose of this research is to analyze sentiment using Convolutional Neural Network and fastText feature expansion. The highest accuracy result is 87.74%.