하이브리드 검색 기반 검색 증강 생성 기술을 활용한 개인 맞춤형 관광 일정 추천 서비스 개발

초록

This paper proposes a personalized travel itinerary recommendation service that leverages search-augmented text generation technology to overcome the factual inaccuracy and lack of real- time information in existing search systems and standalone large language models. When a user inputs a natural language query such as “Plan a 3- day, 2-night summer trip to Tokyo,” the system first performs semantic vector search, which retrieves sentences related to topics such as “travel reviews” or “tourist spot introductions,” alongside keyword-based search that identifies directly matched terms like “Tokyo,” “summer,” or “3-day trip.” The retrieved results are then combined using a score-normalization method to balance their relative importance, and a re-ranking process evaluates the true contextual relevance between the query and the results to prioritize the most appropriate tourism information. Finally, a knowledge graph integrating relational data such as geographic distances and travel times is applied to enhance the logical consistency and feasibility of the itinerary. By integrating these components, this study introduces a multi-stage pipeline combining hybrid search, normalization, re-ranking, and knowledge-graph reasoning, which optimizes accuracy, contextual relevance, and practical feasibility simultaneously. The proposed system generates highly structured itineraries enriched with essential metadata such as geographic coordinates, facilitating seamless integration with map services and enabling dynamic adjustment based on real-time user feedback. This approach significantly enhances the flexibility, reliability, and personalization of travel planning, establishing a foundation for next-generation intelligent tourism platforms