This paper introduces WORLD+, a lightweight, Vision Transformer-based model designed for enhanced scene understanding with a specific focus on accessibility-aware image analysis. Building on the foundational WORLD model, WORLD+ eliminates multimodal complexity by using only visual inputs and integrates an object-driven tagging system to support both scene classification and contextual labeling. A key feature of WORLD+ is its ability to detect accessibility-related elements (e.g., ramps, stairs) and incorporate them into a simple yet effective recommender system. This system enables inclusive suggestions for Points of Interest (POIs) based on visual similarity and mobility needs. Experimental results demonstrate strong classification performance and competitive multi-label tagging, highlighting WORLD+ as a scalable, inclusive solution for real-world applications in e-society contexts.