Joint Classification Model of Topic and Polarity: Finding Satisfaction and Dissatisfaction Factors from Airport Service Review

Abstract

This paper proposes a model developed based on Latent Dirichlet Allocation (LDA). It incorporates both a document dataset and the polarity of the document, for example, a positive and negative evaluation, as input data. In the empirical analysis, it was applied to international airport user reviews, in which the quality of services is evaluated. The results show that the proposed model can classify reviews into topics as effectively as the original topic model, and that its user evaluation forecasting ability is also good. Furthermore, this study examined the automatic generation of a polarity dictionary by the model.