Lisa: a touristic chatbot for Lisbon

Abstract

Abstract As cities continue to attract a growing number of visitors, the development of a tailored chatbot catering to the tourists’ unique needs becomes increasingly valuable. In this paper, we propose a new methodology for developing generative implementations of touristic chatbots and demonstrate its application through a prototype created specifically for Lisbon. Utilizing a web-scraped knowledge base with over 2000 website pages, the chatbot offers recommendations for tourist routes, events and places to visit, provides general information about Lisbon (including transportation) and engages with visitors. The initial evaluation was conducted using synthetic datasets—one for question-answering and another for recommendations—to guide experiments in data preprocessing, exploration of different ChatGPT models and improvements to the Retrieval-Augmented Generation pipeline. A subsequent evaluation was performed with experts, including professionals from Turismo de Portugal, focusing on three major areas (recommendation, information and engagement), and the chatbot achieved very strong results on all use cases. This paper contributes to the literature on chatbot development, emphasizing the benefits of advanced machine learning models in the tourism industry and highlighting the potential of iterative optimization and evaluation based on both synthetic and human survey data for downstream tasks. Likewise, the proposed approach can significantly enhance the tourism experience by offering personalized, engaging and efficient digital interactions, thereby improving overall visitor satisfaction and supporting sustainable destination management.

Topics & Concepts