Aspect-based sentiment analysis of tourists’ revisit intention from tourists’ online reviews on theme parks

Abstract

The investigation of tourists’ revisit intention (TRI) is crucial as it provides tourism destinations’ practitioners (TDP) with insights to revolutionise their customer offerings and enhance their business processes while operating at optimum costs. User-generated content (UGC) is beneficial in offering TDP with an understanding of TRI by utilising aspect-based sentiment analysis. Nevertheless, studies primarily detect aspects based on UGC and utilise nouns for feature building and extraction. Additionally, previous studies that utilise text mining and UGC in the investigation of TRI in the tourism destinations domain are scarce as they are typically the result of surveys. Therefore, to address the gap in the studies and the tourism destinations domain, this study embarks on pre-determined aspects with sentiments detection from UGC. The first objective of this study is to identify TRI through a systematic literature review (SLR), establishing a TRI aspect that are supported by works of literature. Secondly, this study detects the TRI aspects from UGC by using Binary Relevance (BR), with an underlying Multinomial Naive Bayes (MNB) classifier. This study also uses Latent Semantic Analysis (LSA) and Latent Dirichlet Allocation (LDA) to extract the topical meaning of the TRI aspects from UGC, establishing TRI meta-model from the perspective of text mining. Thirdly, this study compares the TRI meta-models from the perspective of text mining against the perspective of SLR by using the Jaccard similarity index coefficient. This study discovers that the text mining meta-model is reasonably similar to the meta-model from the works of literature, with an 83.98% similarity. This signifies that the traditional approach of using surveys in the investigation of TRI in the tourism destinations domain may be augmented using this study’s proposed model.