Aspect-Oriented Review Mining and Recommendation System for Mobile Phones Utilizing RoBERTa-Driven Embeddings

Abstract

Translate Article

The rapid expansion of e-commerce platforms has resulted in a vast repository of mobile phone reviews, creating both opportunities and challenges for personalized recommendation systems. Current approaches often face difficulties in handling high-dimensional textual data, class imbalance, and the semantic ambiguity of customer reviews, leading to suboptimal product recommendations. Traditional collaborative filtering and content-based models struggle to effectively capture nuanced sentiment expressions and contextual dependencies within user feedback. To address these issues, this work presents a RoBERTa-Based mobile phone recommendation system that leverages fine-tuned deep contextual embeddings for robust sentiment classification and preference modeling. The framework employs balanced training data, cross-entropy loss optimization, and cosine similarity to compute user-product relevance, thereby overcoming challenges related to data sparsity and biased sentiment representation. Key features of this approach include the integration of advanced transformer-based language modeling for semantic understanding, a dual-layer filtering mechanism combining sentiment and product attributes, and a scalable architecture designed for high classification accuracy and efficient recommendation generation. Experimental evaluation using benchmark datasets demonstrates superior performance in terms of accuracy, precision, recall, F1-score, and RMSE compared to traditional methods. The proposed methodology ensures more reliable, context-aware recommendations that align closely with individual user preferences, effectively bridging the gap between large-scale review analysis and personalized mobile phone suggestions.