Mining tourist preferences and decision support via tourism-oriented knowledge graph

Abstract

Currently, tourism management research is focused on comprehending the fluctuating tourist preferences and devising targeted development strategies through extensive analysis of heterogenous user-generated contents. However, given the online reviews of attractions involve overabundant mixed and intangible dimensions, the widely-used unsupervised text mining could be incomplete or inaccurate. Furthermore, the existing literature typically restricted to the certain types of attractions within several tourist destinations and origins, can hardly guarantee comprehensive insights. To overcome these limitations, the study proposes a novel knowledge-graph-driven framework, involving the systematic construction as well as the thorough investigation and inference of a tourism-oriented knowledge graph (TKG). Following the ontology of domain expertise, 11,296,716 structured triplets of multifaceted knowledge about 1,174,034 tourists and 20,481 attractions within all 340 city-level destinations across China are extracted from multi-source text corpus by the transferring learning on pre-training language model with 43.64–50.65 % accuracy enhancement. In virtue of TKG, a comprehensive decision-support system can be established, which bifurcates into two distinct modes of knowledge application: symbolic query and distributed reasoning. Through the implementation of multiple spatiotemporal analyses via SPARQL queries on TKG, the distribution regularities of tourist preference, causal interpretations, and their effects on destination development can be progressively detected. Refining the distributed representations of objects by injecting abundant contextual knowledge from TKG can significantly enhance the downstream inferential tasks, such as tourist demand prediction and attraction competitive intelligence.