Tourism Event Knowledge Graph for Attractions Recommendation

Abstract

Attractions recommendation faces a severe data sparsity problem. Introducing the knowledge graph into recommendation can alleviate the sparsity effectively. Existing general KGs focus on static knowledge of attractions, but these public attributes provide less value for users' decision making in tourism. To improve the recommendation performance, it is necessary to introduce more meaningful information which is of greater interest to tourists. We present a novel knowledge graph construction paradigm—Travel Event Knowledge Graph (TEKG), which focuses on describing tourists' behaviors and experiences in tourism events. Taking a travelogue as a travel event, we set the travelogue as the main entity instead of attractions, and introduce the "4W1H" event description framework to organize the information in travelogues effectively. We conduct extensive experiments on a real-world rating dataset of attractions. Experimental results prove that the application of TEKG to recommendation methods can significantly improve the model performance.