Analysis of Attraction Aspects of Tourism Villages in Bali Based on Tourist Perceptions Using a BERT-ABSA Method

Abstract

Aspect-Based Sentiment Analysis (ABSA) provides more informative insights than conventional document-level sentiment analysis by identifying sentiment toward specific aspects of a destination. However, existing BERT-based ABSA models often lack domain-specific knowledge, limiting the interpretability of tourism sentiment analysis. This study proposes a Domain-Aware ABSA framework that integrates a tourism-specific domain lexicon with a BERT-based sentiment classification model to analyze tourist reviews of tourism villages in Bali. The proposed framework organizes sentiment into six tourism-related aspects, namely culture, facility, accessibility, service, atmosphere, and price, enabling more structured and interpretable sentiment analysis. Experimental results demonstrate that the proposed framework achieved an overall classification accuracy of 76%, with the highest performance obtained for the positive sentiment class (Precision = 0.94, Recall = 0.77, and F1-score = 0.84). The sentiment distribution indicates that positive reviews account for 66.7% of the dataset, reflecting generally favorable tourist experiences. Aspect-level analysis further reveals that atmosphere, culture, and service receive the highest positive sentiment, whereas facility and price exhibit relatively higher negative sentiment, identifying infrastructure and cost as priority areas for improvement. These findings demonstrate that integrating a Domain-Aware Lexicon with BERT enhances the interpretability of aspect-level sentiment analysis without modifying the underlying transformer architecture. The proposed framework provides actionable insights for tourism managers and policymakers to support evidence-based decision-making and sustainable tourism village development.