From Natural Language to Structured Queries: An LLM-Powered Chatbot for a Smart Tourism Observatory

Abstract

The accessibility of Smart Tourism Initiatives and Tools (STITs) is a cornerstone of digital transformation in the tourism sector. The EUROSTIT project addresses this through a European Observatory that catalogs these technologies. However, sifting through hundreds of results remains a significant challenge for users. This paper presents a specialized conversational agent that facilitates discovery through a filter-based natural language interface. Departing from computationally intensive fine-tuning, our architecture uses prompt engineering - specifically, one-shot Chain-of-Thought (CoT) reasoning - to generate boolean search queries (BSQs). This approach enables the system to distinguish conversational fillers and other irrelevant terms from domain-specific smart tourism entities, autonomously translating free-text into structured database queries. We describe a dynamic web application with a synchronized user experience where conversational input acts as cumulative filters. The performance of different LLMs is compared through various metrics. We conducted quantitative evaluations to gain insight into the feasibility of using readily available commercial LLMs, instead of resource-heavy model fine-tuning for domain-specific boolean search query generation.