Abstract:
Problem: Businesses increasingly rely on online reviews to guide decisions, yet most existing
sentiment tools are limited. They either analyses a single source (e.g., Google or TripAdvisor)
or provide only retrospective scores, offering little actionable guidance. This gap prevents businesses from accessing unified, real-time intelligence across multiple platforms.
Methodology: This research develops a Business Intelligence prototype that aggregates reviews from Google and TripAdvisor, applies advanced NLP models (TextBlob, TF-IDF, Distil-
BERT), and produces structured outputs including sentiment, emotion distribution, aspect analysis, and LLM-generated recommendations. DistilBERT was benchmarked against baseline models, and the system was exposed through /analyze and /compare APIs that return consolidated frontend-ready JSON.
Results: DistilBERT achieved 86.9% accuracy and a Macro-F1 of 0.76, outperforming Text-
Blob (66.5% / 0.56) and TF-IDF (62.9% / 0.26). Functional and integration testing confirmed
correct operation of API endpoints and UI flows, while non-functional testing demonstrated
stable response times (~6.5s per 200 reviews) and scalability up to 50 concurrent users. The prototype delivers actionable, multi-source insights and establishes a foundation for predictive features and competitor benchmarking in retail and services.
