Explainable AI with Fine-Tuned Large Language Models for Sustainable Cultural Heritage Management: A Public Perception Analysis Approach

The redevelopment of cultural heritage areas, especially in historical urban environments, requires a nuanced understanding of public perceptions to balance preservation with modernization. While Aspect-Based Sentiment Analysis (ABSA) has proven valuable in capturing public emotions towards the built environment, it often overlooks implicit emotional cues and subtle sentiments. This study introduces an advanced framework for Aspect Sentiment Quadruple Prediction (ASQP), employing AI-driven techniques to assess public perceptions of Lijiang Ancient Town, a UNESCO World Heritage site in China. By fine-tuning large models using LoRA-based methods, this research enhances aspect- and emotion-related term recognition, integrating BERT, multi-layer BiLSTM, multi-head self-attention, Convolutional Neural Networks (CNN), and Conditional Random Fields (CRF) for entity recognition and sentiment classification. Interpretability is ensured through SHAP (SHapley Additive exPlanations) analysis. Additionally, implicit sentiment data augmentation using Large Language Models (LLMs) uncovers underlying emotional cues that are not overtly expressed in social media content. Analyzing data from platforms such as Weibo, Dazhong Dianping, and Xiaohongshu (2018–2024), the research uncovers key factors influencing public perception, offering actionable insights for heritage site management, urban planning, and the sustainable preservation of cultural heritage. This methodology provides broader implications for global heritage management and sustainable urban development.