INTELLIGENT ALGORITHMS FOR PERSONALIZED TRAVEL ROUTE PLANNING A GRAPH NEURAL NETWORK AND HYBRID MOVNS-A* APPROACH

Abstract

Personalized travel route planning is a combinatorial optimization problem that must simultaneously satisfy diverse user preferences, budget constraints, time limitations, and geographic factors. Existing approaches either rely on rigid rule-based systems or single-objective optimization methods that fail to capture the complexity of real-world tourism scenarios. This paper proposes an intelligent framework comprising three tightly integrated algorithms: (1) a Graph Neural Network (GNN)-based personalized tourist attraction recommendation algorithm that models user–object interaction graphs to capture latent preference patterns; (2) a hybrid Multi-Objective Variable Neighborhood Search combined with A* pathfinding (MOVNS-A*) algorithm for multi-constraint route optimization, balancing travel time, cost, and user satisfaction simultaneously; and (3) a Stacking Regression ensemble model for predicting post-visit user satisfaction scores. Experimental evaluation on a real-world dataset of 15,420 tourist trajectories from Uzbekistan demonstrates that the proposed framework achieves a Precision@10 of 0.847, an NDCG@10 of 0.831, and a mean route satisfaction prediction error (RMSE) of 0.312 — outperforming baseline methods including collaborative filtering, Dijkstra-based routing, and standalone deep learning approaches by margins of 12–23%. The framework is validated through integration into a working web-based travel planning application.