In the e-commerce ecosystem, negative customer reviews, despite often being a numerical minority, represent the most valuable (axiological) business asset for service improvement. However, this value is frequently obscured by the high volume of positive reviews, creating a significant imbalance in the data. This study aims to design and validate a text mining model that is axiologically focused on extracting critical insights from this "minority voice." We applied the Naive Bayes Classifier (NBC) algorithm, augmented with TF-IDF feature weighting, on a dataset of 6,000 reviews from the 'Famous Florist' store. The epistemic challenge of severe data imbalance (5,432 positive vs. 97 negative) was addressed through the methodological intervention of the Synthetic Minority Over-sampling Technique (SMOTE). The model's validity was assessed using 10-Fold Cross-Validation. The epistemic validation results demonstrated the model's validity, achieving an average accuracy of 90%. Crucially, the model achieved a 99% rate for the negative class. This affirms the model's axiological validity: its ability to reliably identify customer complaints (e.g., 'damaged,' 'packaging') and transform raw data into actionable recommendations for improvement.