Análisis de sentimientos de reseñas relacionadas con el turismo: Un enfoque basado en el uso de modelos generativos

Abstract

Abstract: Aspect-based sentiment analysis (ABSA) has emerged as a tool for understanding user perceptions of a product or entity. In this project, ABSA was applied to the tourism sector to analyze Google Reviews and Reddit posts. The ABSA results aim to provide strategic value to management and promote the analyzed tourist destinations. To process the records (9,569 from Google Reviews and 295 from Reddit), a large-scale language model (LLM) solution was designed. The Gemini 2.5 Flash engine was selected due to its efficiency in zero-shot reasoning tasks and its ability to generate structured outputs from heterogeneous and multilingual datasets, proving more efficient than traditional approaches. To facilitate contextual analysis of the visitor experience, a web prototype was built that allows ABSA to be applied to sets of tourist reviews; the LLMs are used to identify the polarity with which each user rated the product.