Abstract
Tourism managers are increasingly turning to the online sphere to gain relevant customer insights. How-ever, current approaches to analyzing vast and rapidly changing user-generated content (UGC) face several limitations. Supervised approaches require significant effort to provide pre-tagged training data and cannot dynamically identify topics mentioned in UGC. On the other hand, unsupervised approaches typically do not support different abstraction levels or enable a successive refinement of analysis in a drill-down manner, which is often expected as a practical requirement of tourism and destination management. Our research objective is, therefore, to extend current supervised approaches for identifying predefined topics by adopt-ing unsupervised approaches using cluster analysis. The results emphasize that unsupervised approaches can (1) detect non-predefined topics dynamically with an accuracy similar to supervised approaches, thus demonstrating the potential to replace them and avoid the necessity of providing pre-tagged training data. (2) To build a topic hierarchy, unsupervised approaches sense more fine-grained topics as an enhancement of predefined topics on a lower level of abstraction, enabling more powerful drill-down-like analyses. Overall, the proposed extended approach to topic detection promises to support tourism management by meaningfully analyzing the increasing mass of visitors’ online feedback. © 2024 The Author(s).