<![CDATA[The rapid growth of e-commerce platforms like Tokopedia has triggered a massive accumulation of over 65,000 customer reviews, yet it is often accompanied by information pollution in the form of non-informative reviews that hinder consumer decision-making processes. This research aims to extract new knowledge regarding these review characteristics through the implementation of the Knowledge Discovery in Database (KDD) framework, integrating a hybrid K-Means Clustering and Random Forest algorithm. Diverging from conventional classification approaches, this study utilizes K-Means as an exploratory instrument to naturally map six latent topic patterns of reviews based on their textual structure. Experiments were conducted on 35,000 data samples using TF-IDF features enriched by cluster labels as structural predictors. The results indicate that the hybrid model achieves 94.41% accuracy with an F1-score of 0.90 for the non-informative class, showing high stability via 5-Fold Cross-Validation (94.56% ± 0.19%) . The most crucial knowledge discovery is evidenced through SHAP analysis, where the cluster feature ranks 7th out of 1,001 predictor features, confirming that semantic grouping provides a richer structural context than pure lexical features . Furthermore, error analysis reveals specific linguistic challenges such as sarcasm and semantic ambiguity as constraints in automated review detection . This research provides a managerial contribution to e-commerce platforms in enhancing information quality and mitigating information overload issues.]]>
