Utilizing Inter-Annotator Agreement For Effective Fake-Review Detection in E-Commerce

With the exponential growth of e-commerce, the proliferation of fake reviews has emerged as a critical concern for consumers and businesses, especially in beauty products, where reviews are essential in purchasing decisions. Distinguishing between real and fake reviews is essential to ensure consumer trust and satisfaction. This paper proposes a novel approach utilizing the Inter-annotator Agreement (IAA) to classify reviews as fake or real, involving annotators in the labeling process. Specific criteria derived from IAA are introduced to aid consumers in making informed decisions when purchasing products online. Additionally, the study integrates the IndoBERT model with various classifier architectures such as Fully Connected (FC), Long Short-Term Memory (LSTM), and Convolutional Neural Network (CNN). Experimental outcomes highlight the LSTM architecture’s superior performance, achieving notable accuracies of 80.27% in training, 79.83% in validation, and 80.01% in testing phases, surpassing other tested classifiers. This research contributes to the advancement of methodologies for identifying fake reviews in dynamic e-commerce settings and underscores its significance in fortifying trust and dependability in online consumer feedback mechanisms. By enhancing the accuracy of review classification, this study aims to mitigate the adverse impacts of fraudulent reviews, thereby fostering a more transparent and reliable online shopping environment.