UGC-Driven Dynamic Evolution of Public Art Installations via Multimodal Transformer: A Tourist Content Perspective
J. J. Jiang · Advanced Electromagnetics · 2026
Public art installations have long been limited by their inability to perceive and respond in real time to visitor-generated content in situ. To address this limitation, this study proposes a dynamically evolving public art installation system driven by tourist user-generated content and supported by a multimodal Transformer architecture. The system processes heterogeneous image, text, and audio inputs through a hierarchical cross-modal attention mechanism to extract collective affective semantics from visitor-generated signals. A differentiable parameter-mapping network then converts these semantic representations into real-time control actions for installation color, morphology, and rhythmic behavior. The framework also incorporates edge-side deployment, lowlatency communication, and continuous data-stream processing, which are essential for public environments involving dense wireless signal propagation and multimodal sensing. Three field tests were conducted in urban public plazas in China. The results show that tri-modal semantic understanding achieved a fusion accuracy of 91.2%, while end-to-end latency remained below 500 ms. Compared with static installations, the proposed system increased average visitor dwell time by approximately 100.2% and secondary user-generated content publication by 121.8%. During 30 days of continuous operation, system stability reached 99.2%. These results demonstrate the feasibility of integrating multimodal semantic understanding, real-time signal acquisition, and responsive visual control for human-machine co-creative public artworks, and provide an engineering pathway for interactive installations operating in complex wireless and acoustic environments.