To solve the challenge of traditional itinerary planning methods being difficult to effectively integrate user personalized interests, dynamic time budgeting, and spatiotemporal constraints, this study proposes a neural itinerary planning model that integrates time perception and adversarial learning. This model adopts an encoder-decoder architecture, where the encoder integrates the semantic, spatial, and temporal features of interest points through multiple layers of Transformers. The decoder generates sequences through auto-regression and introduces a time-aware mechanism to dynamically maintain the remaining time budget. Meanwhile, adversarial training is utilized to enhance the authenticity and rationality of the generated paths. In the tests conducted in four cities, the research method showed an average accuracy improvement of 5.5%, an average recall improvement of 8.2%, and a comprehensive performance indicator F1 value improvement of 7.3% compared to the most competitive benchmark model LLM Planner. The ablation experiment verified the effectiveness of each core component. The large-scale expansion experiment further proves that the model has good generalization ability, scalability, and system robustness. This model can effectively characterize the spatiotemporal behavioral characteristics of users, enhance the personalization, authenticity, and real-time nature of itinerary planning, and provide technical support for smart tourism and personalized services. A new AI method plans personalized city tours by learning your interests and fitting activities into your exact time budget. The system runs faster and uses less computing power than current tools, making it practical for real-time travel apps. Tests show more accurate and realistic trip suggestions, with future potential to adapt to weather, costs, or group travel.
Anding Huang (Wed,) studied this question.