Tourist Attractions Prediction with Enhanced Location Knowledge Graph

Knowledge graph-based recommendation methods have garnered significant research attention in the field of location prediction due to their ability to enhance semantic and relationship modeling, thus improving recommendation quality and methods. However, the conventional location knowledge graphs, constructed with geographical and functional data related to Points Of Interest (POIs), primarily target scenarios within urban areas. Given the disparities between out-of-town behaviors, like tourism, and in-the-town activities, applying location knowledge graphs to tourist attraction prediction remains an open research task. Moreover, current research often emphasizes predicting locations based on spatio-temporal information while neglecting the inclusion of textual data. This limitation arises from data scarcity and the challenges of capturing text semantics. To this end, we propose a novel framework to construct an enhanced location Knowledge Graph with Tourist Review (KGTR) for predicting tourist attractions. In our framework, we go beyond considering only temporal and spatial aspects and incorporate the descriptive information of POIs visited by tourists. This integration enhances the expressiveness of the knowledge graph and enriches its semantics. We utilize text information to augment the node (locations and users) embedding to improve location predictions. Additionally, we newly collect a real-world tourism dataset including the user reviews of POIs, and evaluate the performance of our framework. The experimental results demonstrate that compared with the state-of-the-art method, our proposal improves the accuracy of tourist attraction prediction by an impressive margin of approximately 26.5 %.