Tourist reviews written in Spanish about Mexican Pueblos Mágicos contain rich narrative structures, cultural references, and strong sentiment imbalance. However, collecting and annotating large, balanced datasets is costly and often impractical. REST-MEX 2026 explores whether synthetic data generated by artificial intelligence can serve as a reliable alternative.

The objective of the task is to design a generative model capable of producing a synthetic corpus of 3,000 tourist reviews that resemble authentic opinions in style, content, and sentiment expression. Each generated review must be original, coherent, and explicitly labeled with a sentiment polarity score from 1 (very negative) to 5 (very positive), assigned by the system itself.

Rather than evaluating the generated texts in isolation, REST-MEX 2026 introduces an indirect evaluation framework. The synthetic reviews produced by each participant are used as the only training data for a sentiment classification model. This classifier is then evaluated on a hidden set of real TripAdvisor reviews from Mexican Pueblos Mágicos.

Under this paradigm, the central challenge is not only to generate fluent or convincing text, but to ensure that the synthetic corpus captures the linguistic diversity, sentiment cues, and distributional properties necessary for a model to generalize to real tourist opinions. Synthetic data that leads to better performance on real reviews is considered more effective.

Through this task, REST-MEX 2026 investigates a key research question:

Can synthetic tourist reviews generated by artificial intelligence meaningfully replace or complement real annotated data for sentiment analysis in Spanish?

By addressing this question, the task advances research on generative language models, synthetic data evaluation, and tourism analytics, while maintaining a strong focus on culturally grounded Mexican destinations