Sentiment and Text Analysis of Airbnb Reviews to Understand Traveller Preferences in Tokyo and Singapore

Abstract

Airbnb named by the name of airbed, breakfast and bed has become a major element of the sharing economy that transforms urban tourism by the means of peer-to-peer accommodation and user-created reviews. In this study, the text mining and natural language processing methods are utilised to analyse and compare the experience of travellers who stayed at Airbnb in Tokyo and Singapore. Airbnb reviews of both cities were gathered and pre-processed according to the standard NLP procedures. The word cloud visualisation based on TF-IDF was implemented to find the terms that were often discussed, whereas a lexicon-based method of sentiment analysis was adopted to determine the overall sentimental distributions. Latent Dirichlet Allocation (LDA) was then used to derive major themes that affect the satisfaction of the guests. The findings display that both cities show a predominantly positive sentiment with the average sentiment score of about 0.93 in Tokyo and 0.91 in Singapore though the neutral sentiments can be more hardly differentiated because of the linguistic vagueness and the imbalance of classes. Three themes were identified as quality of service, convenience of location and room condition. The guests of Tokyo were more concerned about being able to access transportation and being close to it, whereas the guests of Singapore were more concerned about cleanliness, hospitality, and fast check-in. Although the results were limited by the methodological constraints based on the granularity of sentiments, the results indicate the usefulness of TF-IDF and LDA in providing practical recommendations to Airbnb hosts and tourism policymakers.