A scoring approach for detecting fake reviews using MRCS similarity metric enhanced by personalized k-means

Abstract

Online commerce has grown in the digital age, and as a result, consumers now depend more than ever on other consumer feedback to make informed purchasing decisions. However, as the importance of reviews has increased, so has the prevalence of fake ones, which now infiltrate platforms and manipulate users' perceptions. This presents a significant challenge to preserving confidence and integrity in online marketplaces. This study addresses the difficulty of identifying fake reviews by introducing a distinctive methodology that incorporates advanced natural language processing (NLP) tools. By including a new metric, mean review cosine similarity (MRCS), which enhances textual similarity assessment for more accurate detection, we improve the identification process. Additionally, an exaggeration detection technique is included, enhancing the model's capacity to identify deceptive variations in review content. Furthermore, an adaptive clustering method differs from traditional k-means classification through modifying clusters to adjust to the constant evolution of misleading linguistic patterns. Empirical validation on the Yelp labeled dataset demonstrates the approach's accuracy (90%), with high precision (89%), recall (95%), and F1 score (92%), indicating its effectiveness and highlighting areas for further refinement.