Mining Users’ Perceptions Through Sentiment and Emotion Analysis to Address Heritage Conservation Strategies

Monitoring architectural heritage is a crucial step in the planning of proper conservation strategies and resource allocation. Current protocols rely on periodic, though infrequent, expert-led inspections, which assess the state of conservation of heritage assets and inform intervention priorities. However, public perceptions, which may suggest alternative courses of action, are seldom considered. This study proposes an innovative methodology integrating public feedback into heritage monitoring via Natural Language Processing (NLP). The framework, applied to ‘70s heritage sites in Italy’s Marche region, integrates Aspect-Based Sentiment Analysis (ABSA) and Aspect-Based Emotion Analysis (ABEA) to systematically analyze user-generated content, identifying heritage-related aspects and classifying sentiment (positive, negative, neutral) and emotions (e.g., joy, anger) from Google Maps reviews. Heritage-specific targets were first identified in user reviews using spaCy-based tokenization. Sentiment classification (positive, negative, neutral) was performed using a pre-trained Bidirectional Encoder Representations from Transformers (BERT) model, while emotions (joy, anger, sadness, fear) were identified using the FEEL-IT algorithm. User perceptions were effectively retrieved, revealing a generally positive sentiment and joy as the most dominant emotion. This approach enables large-scale monitoring based on continuously updated user feedback, which can be integrated into current monitoring protocols to adopt a more comprehensive decision-making approach.