RoIt-XMASA

The RoIt-XMASA dataset is a large-scale multilingual extension of the Cross-lingual Multi-domain Amazon Sentiment Analysis (XMASA) framework, specifically designed to address the "double shift" of language and domain in sentiment analysis for Italian and Romanian. It features a balanced collection of reviews across three primary domains—books, movies, and music—and includes both labeled samples for supervised learning and a significant unlabeled corpus for semi-supervised domain adaptation. The dataset is characterized by naturalistic, polarized rating distributions and highlights distinct linguistic differences, such as Romanian reviews being approximately 2.4 times longer on average than Italian ones.