Scholarlink

This study presents a multimodal deep learning framework to model discrepancies between Tourist-Generated Text (TGT) and Tourist-Generated Photography (TGP) in the construction of destination images. Leveraging semantic network analysis and visual clustering based on Google Vision API and K-Means, the framework extracts cognitive dimensions from textual narratives and affective patterns from visual data. Unlike prior research that emphasizes modality convergence, this study focuses on the asymmetry between text and image in conveying destination perception. Our approach enables scalable, fine-grained modeling of consumer sentiment and symbolic representation across modalities. Experimental results from a dataset of 3.35 million words and 48,000 images reveal that text tends to deliver evaluative and historical content, while images encode emotional appeal and visual atmosphere. The findings not only enrich theoretical understanding of multimodal perception but also inform AI-enhanced strategies for destination branding, content optimization, and consumer behavior prediction in digital tourism environments.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Multimodal Perception Modeling of Tourist Destination Images via Deep Visual Tagging and Semantic Text Graphs

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Shangqing Liu,
  • Ping Yin,
  • Liying Wang

摘要

This study presents a multimodal deep learning framework to model discrepancies between Tourist-Generated Text (TGT) and Tourist-Generated Photography (TGP) in the construction of destination images. Leveraging semantic network analysis and visual clustering based on Google Vision API and K-Means, the framework extracts cognitive dimensions from textual narratives and affective patterns from visual data. Unlike prior research that emphasizes modality convergence, this study focuses on the asymmetry between text and image in conveying destination perception. Our approach enables scalable, fine-grained modeling of consumer sentiment and symbolic representation across modalities. Experimental results from a dataset of 3.35 million words and 48,000 images reveal that text tends to deliver evaluative and historical content, while images encode emotional appeal and visual atmosphere. The findings not only enrich theoretical understanding of multimodal perception but also inform AI-enhanced strategies for destination branding, content optimization, and consumer behavior prediction in digital tourism environments.