Authors: Andrei Kirilenko*, University of Florida, Svetlana Stepchenkova, University of Florida
Topics: Tourism Geography, Communication, Quantitative Methods
Keywords: Tourist satisfaction, Latent Dirichlet allocation, TripAdvisor, Social media
Session Type: Paper
Start / End Time: 12:30 PM / 1:45 PM
Room: Virtual Track 11
Presentation File: No File Uploaded
Automated content analysis of online travel reviews allows analysis of topics of travelers’ satisfaction and dissatisfaction, yet its domain is not well researched. We suggest that the “Anna Karenina principle” positing a greater variability of the factors leading to business failure as opposed to those leading to its success results in limitations of topic modeling applied to dissatisfied visitor reviews. We test our hypothesis using TripAdvisor reviews of several locations worldwide, in multiple languages. Methodologically, the reviews are translated, cleaned, and then the topics are extracted with Latent Dirichlet allocation using the coherence measure to determine the optimal number of topics. The results confirm our initial hypothesis that the topical on the low customer satisfaction datasets are significantly less interpretable compared with the analysis of the positive reviews. We suggest that this effect is due to the observed diversity of negative opinions leading to a much greater diversity of the negative comments as compared with the positive ones. If confirmed, it would mean that the Anna Karenina effect limits application of the automated topic modeling to the analysis of the main topics of customer dissatisfaction to very large datasets where the sheer volume of reviews would warrant keeping much greater number of topics arising from topical analysis.