The emergence of online commerce in the digital age resulted to an unparalleled dependence on consumer feedback for making informed choices. However, the rise in the importance of reviews has coincided with an increase in fraudulent behavior, as fake reviews have come into platforms and influenced users' opinions. The widespread use of these deceptive reviews has become an increasing issue, reducing the trust that is essential for the success of online marketplaces.
This study uses a comprehensive approach to handle the difficulty of identifying fake reviews. This distinctive strategy introducing a multidimensional methodology incorporating advanced Natural Language Processing (NLP) tools. The utilization of Word2Vec for semantic richness in review analysis is detailed, emphasizing its ability to capture contextual complexities. The integration of new metric navigates the semantic landscape, offering a normalized measure of textual closeness critical for precise detection. A distinctive feature is the incorporation of an exaggeration detection method, which improves the model's capacity to reveal misleading complexities in reviews' language. Moreover, an adaptive clustering is elaborated, which deviates from conventional K-means clustering by continuously adjusting clusters to account for the dynamic nature of deceptive language patterns.
The approach’s excellent accuracy is demonstrated via empirical validation on the Yelp Labeled Dataset, which also highlights areas for development. Metrics like precision, recall, and F1 score provide light on the model's advantages and direct further improvements. The conclusion examines how misleading reviews affect digital trust more broadly and highlights the proposed approach as a critical first step in reducing fraudulent activity.