Consumer Cynicism Identification for Spanish Reviews using a Spanish Transformer Model

Abstract

Companies pay close attention to how consumers react on social media to their products or services. Our work focuses on the identification of Consumer Cynicism, defined as a negative attitude that can have a broad or specific focus and comprises cognitive, affective, and behavioral components. We create a corpus of 619 Spanish-language comments on YouTube car reviews, annotated for four cynicism constructs: Dissatisfaction, Alienation, Skepticism, and Hostility. We compare different classification formulations (binary vs. multi-label) and different pre-trained models (Spanish BETO vs. multilingual BERT). We find binary classifiers derived from BETO consistently outperform multi-label classifiers and classifiers derived from BERT. Our best models achieve F1 of 0.83 for Dissatisfaction, 0.77 for Hostility, 0.71 for Skepticism and 0.70 for Alienation.