Sistem Ekstraksi Keluhan Tempat Wisata dari Komentar Google-Maps Menggunakan IndoBERT dan POS Tagging

Abstract

Google Maps comments are not only used by tourists as references in selecting destinations, but also reflect visitor satisfaction and complaints. Unaddressed complaints may reduce destination ratings, influence other tourists’ decisions, and harm the image of tourist attractions. However, such comments are unstructured textual data, requiring specific analytical methods to be processed effectively. This study aims to overcome the limitations of general online comment analysis by applying the IndoBERT model for sentiment analysis and combining it with Part of-Speech (POS) Tagging to extract more specific complaint focuses. The data were collected through scraping Google Maps comments from 20 tourist attractions in Jember Regency. The analysis process included clause boundary detection, manual sentiment labeling into negative and non-negative classes, preprocessing, data splitting, word embedding construction, fine-tuning the IndoBERT model, and sentiment evaluation. The evaluation results show that the proposed model achieved an accuracy of 97.93% and a recall of 99.24%. Furthermore, comments classified as negative were processed for complaint extraction using POS Tagging. The developed model was implemented in a web-based application. The results indicate that while the model effectively distinguishes negative and non-negative comments, the complaint extraction feature still has limitations due to the rule based POS Tagging approach, which struggles to capture linguistic variations and contextual complexity.