SentimentLens analyzes 10K hotel reviews using aspect-based AI to resolve rating conflicts

SentimentLens, developed by Dineth Jayakody, Pasindu Thenahandi, and Sampath Jayarathna, is a scalable AI framework designed to transform unstructured hotel reviews into structured, actionable intelligence. It leverages Aspect-Based Sentiment Analysis (ABSA) to break down reviews into specific aspects (e.g., cleanliness, staff, location), classify sentiment for each, and assign semantic categories. The system then aggregates these insights at three levels: region, hotel, and service category, enabling granular comparisons across geographic contexts and hospitality settings.

The standout feature is its cross-modal reconciliation module, which aligns textual sentiment with numerical star ratings to uncover latent conflicts—e.g., a 4-star hotel with consistently negative reviews about staff. Using importance-performance analysis and entropy-based metrics, SentimentLens identifies high-impact improvement opportunities and structural inconsistencies in service quality. Demonstrated on a dataset of over 10,000 public hotel reviews, the framework proved effective in revealing how traveler sentiment varies across regions and hotel archetypes. The approach is designed to be generalizable to other review-driven sectors like restaurants, airlines, and e-commerce.