Mobile Spatial Informatics: Bridging "Regional Memory" and "Human Experience"
"Why is this road winding?" "Why are there so many temples on this street?" Every local landscape is built upon overlapping layers of history and the rationale of the people who lived there. However, in today's era of urbanization and population decline, the storytellers who can articulate these layers are disappearing, making the meanings behind the scenery invisible to us. "Regional Memory" fails to reach the "Walking Person." Bridging this disconnect is our starting point. From the perspectives of Spatial Informatics and Mobile Computing, this research aims to develop and implement technologies that re-evaluate local diversity, charm, and civic pride. By leveraging AI and Data Science to aggregate local history and wisdom, we explore methodologies to build a "Human-Centric Information Infrastructure" that fosters collaboration among residents, visitors, and the community to revitalize the region.
Walking tourism is gaining popularity across the world, but evaluating whether visitors are truly satisfied with a planned route is difficult. Questions such as “Where do tourists stop?” or “Which places do they pass without interest?” are hard to answer objectively. Traditional surveys require significant cost and effort, making continuous improvement of tourist routes challenging.
📊 Mobile Sensing for Context-Aware Tourism Analytics
This study leverages sensor data automatically collected by smartphones to build a sustainable improvement cycle for tourism content. By combining GPS traces with accelerometer data (walking vs. stopping), GPS accuracy information (indoor vs. outdoor estimation), and app interaction logs (photo-taking events), our system automatically infers the context of tourist behavior—where visitors are and what they are doing. Using these inferred contexts, we generate density maps to analyze overall movement patterns and provide quantitative evaluation scores that measure how well actual tourist behavior aligns with planned routes. This enables continuous, data-driven refinement of walking tourism experiences.
Many navigation and sightseeing apps automatically play audio such as “You are now near X” when you approach a landmark. This push-based location service relies on geofencing, which triggers content when users cross virtual boundaries on a map. Designing geofences, however, is challenging: GPS accuracy varies, tourists move unpredictably, and service goals differ. Manual design cannot fully accommodate these factors.
📍 Data-Driven Optimization for Tourist Guidance
In this study, we propose a data-driven approach to automatically generate geofence parameters using tourist behavior data. Machine learning techniques identify how users interact with sightseeing spots—such as entering a building versus viewing it from outside—while evolutionary computation refines geofence placement to trigger audio where many users naturally pass. This combination enables adaptive, efficient, and context-aware geofence design, improving the reliability and relevance of location-based audio guides across diverse tourism scenarios.
Kevin Lynch, an American urban planner, pointed out the importance of the "city image," stating that it is vital for people to recognize the identity, structure, and meaning of the urban environment so that it remains a vivid memory. For both visitors and residents, learning about local history, culture, and episodes is a crucial element in enhancing the quality of the regional experience. While smartphone-based audio tours have become widespread in recent years, many services only play guidance upon arrival at specific spots, leaving users walking in silence between them. Traditional Point-of-Interest (POI) based tours often lack sufficient information in areas with few conspicuous spots, resulting in silence for the majority of the walk. Furthermore, deviating from the planned route can often disrupt the consistency of the story. Can such fragmented regional guides truly foster a coherent regional image?
🎙️ Local Storytelling Intelligence where Walking Itself Becomes a Narrative
In this research, we are developing "Onsite Radio AI," a system that transforms every moment of a walk into storytelling by combining Hierarchical Geofencing and Large Language Models (LLMs). Hierarchical Geofencing is a location recognition algorithm that simultaneously handles multiple spatial scales, such as entire districts, streets, and individual buildings. Much like a tour guide who explains specific landmarks when they are nearby but switches to discussing the town's general history when there are no conspicuous objects around, this system can smoothly switch the geographic scale of the topic according to the listener's current location. Furthermore, using historical and cultural archives accumulated in the region, the LLM instantly generates natural, conversational guidance based on the recognition results of that geographic scale. We are researching this type of local storytelling intelligence that cultivates the "Imageability" of a region derived from the user's movement trajectory and environmental changes.