Aspect-Based Sentiment Analysis Using Latent Dirichlet Allocation (LDA) and DistilBERT on Threads App Reviews

Threads is a social media application that offers news services and user interaction, integrated with Instagram. Unlike other platforms, Threads does not have features like direct messaging (DM), trending topics, or advertisements. To understand users' opinions about this app, a sentiment analysis based on aspects was conducted on Threads reviews. The steps involved include applying web scraping techniques to collect reviews data from the Play Store. Aspect categories were identified using the Latent Dirichlet Allocation (LDA) algorithm. Sentiment labeling was then performed for positive and negative categories using the DistilBERT method. The results show that the LDA algorithm identified three aspects: Usage and Integration (with 3.147 positive and 8.173 negative reviews), Features and Comparisons (with 1.108 positive and 1.709 negative reviews), and User Experience and Satisfaction (with 3.529 positive and 2.208 negative reviews). The sentiment analysis results indicated 7,784 positive reviews and 12,090 negative reviews. Model performance evaluation using the Confusion Matrix showed an accuracy of 96.71%, precision of 97.24%, recall of 94.48%, and F1-score of 95.84%. Evaluation was also conducted for each aspect, with an accuracy of 96.99%, precision of 96.60%, recall of 92.85%, and F1-score of 94.69% for the Usage and Integration aspect; accuracy of 95.74%, precision of 94.11%, recall of 95.23%, and F1-score of 94.67% for the Features and Comparisons aspect; and accuracy of 96.74%, precision of 95.83%, recall of 99.06%, and F1-score of 97.42% for the User Experience and Satisfaction aspect.