Prototipo de información turística basado en GraphRAG. Caso de aplicación: Recomendación de destinos turísticos del Ecuador

Abstract

Abstract: This research addresses one of the problems faced by today's tourism platforms: information overload and a lack of personalization. These platforms tend to offer general recommendations without considering the user's situation. In response to this problem, the idea arose to create a tourism recommendation system, PlaceRed, designed to boost tourism in the province of Loja, Ecuador. The proposal is based on the hybrid architecture of Graph Augmented Retrieval (GraphRAG). PlaceRed attempts to overcome one of the limitations of Long-Term Learning Models (LLMs): the hallucinations these models can produce. PlaceRed draws on a knowledge graph stored on Neo4j (vector and relational storage) and uses a cascaded LLM orchestration strategy (Gemini 1.5 Flash, DeepSeek, and Ollama) executed on Node.js as the server backend. PlaceRed offers users semantic search and structured search functionalities based on filters, allowing them to find specific information according to their requirements. The results show that the hybrid solution, through a reactive interface, allows for answering complex queries in natural language, offering contextualized, explained and visually enriched recommendations, which can be enriching for the tourist.