Visitor feedback is a critical yet underutilized resource for optimizing cultural heritage experiences. This study introduces a novel hybrid sentiment analysis framework that integrates digital reviews (Google, TripAdvisor, Traveloka) with digitized physical guestbook entries, capturing both reactive online sentiment and reflective on-site feedback. Leveraging a Convolutional Neural Network (CNN) with custom-trained Word2Vec embeddings, our model classifies visitor sentiment into Positive, Neutral, and Negative categories with 94.7% accuracy and 0.93 F1-score, significantly outperforming traditional models (SVM, Naïve Bayes). Analysis reveals a striking sentiment divergence: physical guestbooks exhibit 71.2% positivity, compared to 58.9% on Google Reviews, highlighting a systematic negativity bias in digital platforms. We further prototype a real-time dashboard for museum staff, enabling data-driven interventions based on sentiment drift and keyword trends. This work pioneers the fusion of analog and digital visitor voices in Indonesian cultural analytics, offering museums a scalable, explainable, and context-aware tool for experience optimization.
Fauziah et al. (Wed,) studied this question.