Deep Learning for Multi-Aspect Sentiment Analysis of TikTok App using the RNN-LSTM Method

Abstract

Applications built expressly for consumers to communicate online are known as social media apps. Social media applications are utilized for enjoyment as well as for interacting. For Android users, applications may be found in the Google Play Store, while for iOS users, they can be found in the Apple App Store. The site offers a collection that is a big resource-rich in thoughts, opinions, and feelings, notably on Google Playstore. Each user's review has an aspect value. Due to a large number of reviews, sentiment analysis is tough. The author proposes to do an Aspect-Based Sentiment Analysis (ABSA) utilizing TikTok app reviews on the Google Play Store in this paper. Currently, there are 65.2 million active users of the Tik Tok program, including 8.5 million users from Indonesia, there are still a few studies that use the TikTok application dataset. In this study, sentiment classification is carried out on each aspect that has been determined, namely, aspects of features, business, and content, the method used is deep learning Recurrent Neural Network with the Long Short-Term Memory (RNN – LSTM) model and the addition of word embedding BERT. The results showed that the classification of sentiment in the business aspect showed the highest score, namely 0.94, the sentiment classification in the aspect received an accuracy of 0.91 while the feature aspect got the lowest accuracy, which was 0.85.