Overview of Rest-Mex at IberLEF 2025: Researching Sentiment Evaluation in Text for Mexican Magical Towns

Resumen

The REST-MEX 2025 shared task, hosted at IberLEF, represents the fourth edition of an ongoing effort to benchmark natural language processing systems in the domain of Mexican tourism. The task focuses on the automatic classification of user-generated reviews into three axes: polarity (from 1 to 5), type of service (hotel, restaurant, or attraction), and the identification of one of 40 predefined Pueblos Mágicos (Magical Towns). This year, the challenge attracted a record number of submissions and introduced greater complexity by expanding the dataset and encouraging research into advanced fine-tuning strategies, prompt engineering, and data-centric methods. A total of 32 teams submitted 70 valid runs, exploring a diverse range of methods including Transformer-based models, transfer learning from previous REST-MEX editions, class binarization schemes, and instance selection strategies based on contextual embeddings. This paper presents an overview of the task, the dataset characteristics, the evaluation protocol, and a comparative analysis of the results.