Dynamic prediction and recommendation of museum visitors' interest based on long short-term memory network (LSTM)

Abstract

In the era of personalized digital experiences, enhancing visitor engagement in museums through intelligent, adaptive recommendation systems has become a critical need. Traditional museum recommendation systems often fail to account for the dynamic and evolving nature of visitors' interests, which may shift during a single visit or across multiple visits. Hence, personalized, time-aware models that adapt to changing behaviors and deliver contextually relevant recommendations are demanded. For this, a Dynamic Attention-based LSTM for Interest Recommendation (DALIR), a sequential deep learning framework, is designed to model and predict the evolving preferences of museum visitors. Using an LSTM architecture enhanced with a dynamic attention module, DALIR focuses on the most influential aspects of a visitor's behavior, such as dwell time and exhibit sequence, to generate accurate and timely recommendations. The model is trained on a rich dataset that captures sequential interaction patterns, including spatial transitions and engagement intensity, enabling personalized guidance throughout the museum experience. DALIR dynamically updates its predictions based on recent visitor actions, ensuring contextual relevance and adaptive content delivery. DALIR achieves a 15.9% improvement in visit sequence prediction accuracy and a 16.6% improvement in dwell time prediction accuracy over baseline models (TMSNN, GCN, LDA-LSTM). Additionally, DALIR enhances precision by 18% for visit sequences and 11% for dwell times, while improving recall by 16.7% and 9.9%, respectively. These results demonstrate the effectiveness of DALIR’s dynamic attention mechanism and sequence-adaptive modeling in capturing evolving visitor interests. DALIR provides a scalable, intelligent recommendation solution for curators and museum platforms, enhancing visitor satisfaction and content discovery through interest-driven experiences.