UGuideRAG: Intent-Enhanced Retrieval-Augmented UGuideRAG: Intent-Enhanced Retrieval-Augmented Generation with User-Generated Content for Personalized Urban Tourism Generation with User-Generated Content for Personalized Urban Tourism

Citywalk, as an increasingly popular form of urban tourism, emphasizes immersive, diverse, and personalized exploration over conventional sightseeing. These features evolving tourist expectations pose new challenges for intelligent itinerary planning, particularly in capturing the rich experiential attributes of visitor attractions and aligning them with ambiguous and underspecified natural language queries. This thesis proposes UGuideRAG (User-Generated Content-Guided Retrieval-Augmented Generation), a modular framework that leverages user-generated content to con struct a comprehensive attraction database, employs large language models for intent-enhanced retrieval and recommendation, and incorporates spatial optimiza tion to ensure coherent itinerary planning. By bridging the gap between par tially expressed user goals and the multi-dimensional nature of urban experiences, UGuideRAG enables more insightful and personalized trip recommendations. For walk-centric route planning, UGuideRAG further constructs a scenic pathway database by fusing POI data with geotagged photos to estimate segment-level scenicness using photo density and street interactivity, and integrates this score into a multi-objective route generator that links the candidate attractions while balancing distance, spa tial compactness, and accumulated scenic value. Experiments on real-world datasets demonstrate that the proposed framework consistently surpasses existing methods in producing contextually relevant, user-centered, and spatially optimized urban tourism itineraries.