Enhancing Recommendation Systems via Aspect-Level Sentiment Analysis for Identifying Unreliable Reviews

El Mehdi Lghaouch (Laboratory of Information Technology and Modeling, Faculty of Sciences Ben M’Sik, Hassan II University, Casablanca)
Fatima Zahra Abbour (Laboratory of Information Technology and Modeling, Faculty of Sciences Ben M’Sik, Hassan II University, Casablanca)
Soumaya Ounacer (Laboratory of Information Technology and Modeling, Faculty of Sciences Ben M’Sik, Hassan II University, Casablanca)
Soufiane Ardchir (National School of Management and Marketing, Hassan II University, Casablanca)
Mohamed Azzouazi (Laboratory of Information Technology and Modeling, Faculty of Sciences Ben M’Sik, Hassan II University, Casablanca)

Indonesian Journal of Electrical Engineering and Informatics (IJEEI)
Vol 14, No 2: June 2026

User-generated reviews are an important source of information for e-commerce recommendation systems. However, reviews may contain conflicting opinions across product aspects such as quality, price, and delivery, which can weaken both review-level trust and aspect-level signals. To address this issue, we propose AspectGuard, a filtering framework based on Aspect-Based Sentiment Analysis (ABSA) and a Sentiment Divergence Score (SDS) that quantifies disagreement across aspect-level sentiments within a review. Reviews with high divergence are identified as potentially unreliable and excluded before constructing recommendation profiles. To isolate the contribution of the proposed filtering mechanism, we compare the same content-based recommendation model with and without AspectGuard filtering. Experiments conducted on a subset of the Amazon Reviews 2023 Electronics dataset show that the proposed filtering strategy improves both rating prediction and ranking quality. These results indicate that aspect level sentiment divergence can serve as an effective signal for identifying unreliable reviews and improving the reliability of recommendation systems.