This study explores a critical feedback loop of algorithmic bias on e-commerce platforms, where the design of recommendation systems incentivizes user behaviors that ultimately distort the data these systems depend on. Specifically, algorithms that prioritize products with higher ratings encourage users to submit inauthentic 5-star reviews—even after negative experiences—in order to boost the visibility of their feedback. This tactic manipulates algorithmic sorting, creating a self-reinforcing cycle that artificially inflates product scores and misleads consumers. An exploratory analysis was conducted and developed a phrase-based polarity dataset to uncover sentiment-rating mismatches. We evaluated a range of natural language processing techniques, including supervised classifiers, unsupervised methods, and Turkish-specific sentiment corpora. These approaches were validated against a manually annotated subset of reviews. The findings offer a robust framework for detecting this pervasive form of platform manipulation and underscore the urgent need to design recommendation algorithms that are resilient to strategically motivated bias. Clustering enabled the identification of latent emotional patterns across reviews, particularly those expressing negative sentiment masked by high ratings.