Coupling space features of transformer digital signage advertising theme recommendation

This study coupled spatial features with the Transformer model to conduct in-depth research on the extraction and recommendation methods of digital signage advertisement themes, focusing on addressing the issue that existing digital signage recommendation research fails to integrate spatial location and advertising topic, two crucial factors for precise recommendations. The research crawled digital signage advertisement images and POI data within the Sixth Ring Road of Beijing as the study dataset, designed the RPRSTR model for identifying text content in advertisement images, employed the HSBERT model for the classification of advertising topics, and utilized the ML-TRec model for the recommendation of advertising topic types. Experimental results indicate that the RPRSTR model maintains a high text recognition accuracy rate against complex backgrounds; the HSBERT model effectively reduces computational complexity when dealing with large output vocabulary tables and enhances the accuracy of topic classification; the ML-TRec model demonstrates superior classification performance across multiple evaluation metrics. This method is expected to facilitate the scientific planning and marketing efficiency of urban outdoor digital signage and provides a new application paradigm for the use of geospatial big data and spatial intelligence technology in the digital economy.