Quantifying Review Informativeness with Aspect-Aware Knowledge Graphs and Linguistic Parsing

The increasing reliance on user reviews for decision-making has elevated the role of online reviews on e-commerce platforms, yet the surge in generic, vague, or promotional content undermines their informativeness and credibility. While existing sentiment analysis and fake review detection methods classify reviews based on polarity or authenticity, they often overlook a critical dimension: how informative and actionable a review truly is. Reviews such as “Great place!” may indicate positive sentiment but lack specific, aspect-level insights needed by consumers or businesses. This deficiency creates a substantial gap in current review analysis systems, which struggle to differentiate between superficial and detailed feedback. Addressing this gap requires an approach capable of capturing both the presence and richness of aspect-specific content. This paper presents a novel knowledge-driven approach for evaluating the informativeness of restaurant reviews by integrating linguistic analysis with Knowledge Graph (KG) representation. The KG is constructed using restaurant reviews from the Yelp dataset to define aspect categories, extract aspect terms, and model their corresponding sentiment relations. The proposed approach systematically identifies and extracts aspect-sentiment pairs related to four critical aspects: food, service, price, and atmosphere, using dependency parsing and a curated lexicon. A custom metric, Aspect-Aware Length-Normalized Information Density (Aspect-Aware LNID), is introduced to quantify the richness and completeness of a review based on its semantic content and aspect coverage. Experimental evaluation on the SemEval 2014 dataset demonstrates the effectiveness of the proposed approach, achieving a 97.5% weighted-average F1-score in aspect category extraction and providing interpretable, scalable, and domain-specific insights. This hybrid approach offers a transparent alternative to opaque deep learning models, enabling fine-grained opinion mining and improved consumer trust in online review platforms.