Transformers (BERT) Based Framework for Web Recommendations Using Sentiment-Enriched Web Data

Abstract

Web mining and Natural Language Processing (NLP) play a crucial role in modern recommendation systems, enhancing accuracy and personalization by leveraging user-generated content. This research proposes a novel framework integrating Transformers, specifically BERT (Bidirectional Encoder Representations from Transformers), to detect fake reviews and perform sentiment analysis. Traditional recommendation techniques, such as collaborative and content-based filtering, fail to capture nuanced user sentiments, leading to suboptimal results. The proposed model refines web-based recommendations by filtering out fake reviews and extracting sentiment-enriched insights, ensuring more reliable predictions. The system undergoes extensive evaluation using machine learning algorithms, including K-Nearest Neighbors, Multinomial Naïve Bayes, Logistic Regression, Random Forest, AdaBoost, and XGBoost, with BERT demonstrating superior performance. Experimental results highlight an impressive accuracy of 94.34% for fake review detection and 94.78% for sentiment classification, outperforming conventional models. The integration of sentiment-driven web mining enhances recommendation accuracy, mitigates misleading feedback, and improves user trust. This study underscores the potential of Transformer -powered sentiment analysis in refining recommendation systems for e-commerce and other digital platforms.