Deep learning-based research on cultural symbols in few-shot destination social media images

Abstract

In the digital age, cultural heritage sites utilise social media imagery for marketing. Nevertheless, existing research reveals two gaps: the scarcity of methods for accurately identifying multi-cultural element fusion in images under few-shot conditions; and the difficulty in quantifying their collective impact on user behaviour, which limits data-driven content strategy optimisation. Grounded in destination image theory and symbolic consumption literature, this study selects Shanghai’s Yuyuan Tourist Mart as a case study. It proposes GNN-CNN-CLIP, a deep learning framework integrating Graph Neural Network (GNN), Convolutional Neural Network (CNN), and Contrastive Language-Image Pre-training (CLIP) models to automatically identify six cultural element categories in images. Cluster analysis then clarifies their links to user interaction preferences. A deep learning analysis of 500 promotional images shows that narrative content integrating both material and behavioural cultural symbols drives engagement more effectively than material symbols alone. User interaction clustering reveals three distinct groups, providing critical insights for precision content marketing at cultural heritage sites. The proposed deep learning framework offers a solution for few-shot image analysis, enabling data-driven optimisation and segmented strategies for cultural heritage sites. These findings extend destination image formation theory by demonstrating that behavioural cultural symbols transform static destination representations into dynamic narratives resonating with digital audiences.