Extracting principal smartness dimensions of smart speakers using topic modeling and sentiment analysis

Abstract

Although the smart speaker market has experienced massive growth in recent years, there is a lack of research on what consumers really consider important for so called 'smart' speakers, which are supposedly distinguished from traditional products. Therefore, this study aims at identifying key smartness dimensions that are related to the satisfaction of smart speaker users. First, a total of seven topics were extracted from the Amazon's review data through Latent Dirichlet Allocation, and the topics were mapped, through group discussion, to three smartness elements defined from the literature review. Then, sentiment scores of each topic were calculated using SentiWordNet, which were then used as variables to develop star rating classifiers. The feature importance of the classifiers revealed that the connectivity issue is the most influential factor in determining the customer satisfaction of smart speakers. The next important topics are sound quality and its use as a media player. The study findings have direct practical implications on smart speaker development.