Title: Enhancing Restaurant Feedback Interpretation Using a Knowledge-Based System Integrated with NLP Techniques
Authors: Mohammed khair I. Kassab , Samy S. Abu-Naser
Volume: 9
Issue: 8
Pages: 121-128
Publication Date: 2025/08/28
Abstract:
This study introduces a hybrid sentiment analysis framework that integrates transformer-based Natural Language Processing (NLP) with a symbolic rule-based Knowledge-Based System (KBS) to enhance the interpretation of restaurant customer feedback. The architecture combines a fine-tuned BERT model for sentence-level sentiment classification with a CLIPS-based inference engine to ensure contextual accuracy and decision transparency. The NLP component processes unstructured reviews, extracting aspect-specific sentiments across food, service, hygiene, pricing, and customer loyalty. The KBS component refines predictions using domain-specific rules, particularly for context-sensitive expressions-such as interpreting "cheap" as negative in pricing contexts. The system was implemented using Python and Hugging Face Transformers and evaluated on real-world datasets from platforms like Yelp and TripAdvisor. Experimental results demonstrate that the hybrid model achieved an average accuracy of 89% and an F1-score of 0.86, outperforming traditional machine learning baselines. It also significantly improved interpretability, especially in low-confidence cases, and offered traceable reasoning for managerial decision support. This research underscores the value of neuro-symbolic integration in real-time sentiment analysis and provides a practical foundation for extending hybrid frameworks to multilingual, adaptive, and cross-domain applications.