CuPe-KG: Cultural perspective-based knowledge graph construction of tourism resources via pretrained language models

Tourism knowledge graphs lack cultural content, limiting their usefulness for cultural tourists. This paper presents the development of a cultural perspective-based knowledge graph (CuPe-KG). We evaluated fine-tuning ERNIE 3.0 (FT-ERNIE) and ChatGPT for cultural type recognition to strengthen the relationship between tourism resources and cultures. Our investigation used an annotated cultural tourism resource dataset containing 2,745 items across 16 cultural types. The results showed accuracy scores for FT-ERNIE and ChatGPT of 0.81 and 0.12, respectively, with FT-ERNIE achieving a micro-F1 score of 0.93, a 26 percentage point lead over ChatGPT's score of 0.67. These underscore FT-ERNIE's superior performance (the shortcoming is the need to annotate data) while highlighting ChatGPT's limitations because of insufficient Chinese training data and lower identification accuracy in professional knowledge. A novel ontology was designed to facilitate the construction of CuPe-KG, including elements such as cultural types, historical figures, events, and intangible cultural heritage. CuPe-KG effectively addresses cultural tourism visitors' information retrieval needs.