ADVANCED HYBRID INTELLIGENT RECOMMENDER SYSTEMS AND DATA ENGINEERING LIFECYCLE FOR OPTIMIZING MULTI-SIDED B2B2C ECOSYSTEMS IN DIGITAL TOURISM

Abstract

This research addresses the critical problem of operational friction and data sparsity in multi-sided digital tourism marketplaces operating under the hybrid Business-to-Business-to-Consumer (B2B2C) framework. As platforms scale, information overload and the asymmetric visibility of micro-operators present severe structural challenges. Using the architectural blueprint of Safarli.uz, we design, formally specify, and empirically evaluate an advanced hybrid intelligent recommender system. The framework seamlessly unifies spatial-semantic content-based filtering with low-rank latent factor collaborative filtering via Singular Value Decomposition (SVD). To resolve the structural bias against newly onboarded B2B micro-operators, we introduce a mathematical model driven by a Dynamic Novelty Boosting decay algorithm. This model ensures cold-start items receive targeted algorithmic exposure to optimize implicit telemetry gathering. Simulation experiments carried out on synthetic and historical data streams demonstrate that the proposed hybrid architecture achieves substantial performance gains, increasing Precision@5 by 216.6% and shifting the Click-Through Rate (CTR) from 4.2% to 14.5%, while successfully neutralizing the cold-start structural constraint.