Abstract
To address the challenges associated with the fragmented distribution of scenic attractions, the ever-evolving nature of user preferences, and the suboptimal performance of multi-objective optimisation in rural tourism settings, this research introduces an innovative route recommendation approach that synergistically combines the greedy algorithm with an attention mechanism. The methodology involves the development of a dual-channel attention framework, which dynamically assigns weights to user behaviour patterns and textual semantic features. Additionally, a multi-constraint path-generation architecture, grounded in the greedy algorithm, is engineered to harmonise conflicting objectives, including time expenditure, alignment with attraction preferences, travel distance, and financial limitations. The results show that when recommending 20 scenic spots, the average absolute error decreases to 0.6921, the running time is only 0.1762 seconds, the coverage of scenic spot types increases to 6.5, and the recommended diversity index increases to 0.85. Compared with mainstream recommendation algorithms, this method significantly leads in total route revenue, accuracy of 0.83, and F1 score of 0.80, and has the lowest computation time. Application analysis shows that the recommended route has a high coverage rate, and the time consumption slows down with the increase of the number of scenic spots, taking into account personalised needs and diversity of scenic spot types. The method proposed by the research can effectively balance the real-time and personalised needs of route planning, while reducing path redundancy and enhancing the interpretability of recommendations, providing a new technological path for rural tourism route recommendations.