Ranking online retailers using unsupervised machine learning

Online reviews help customers make an informed purchasing decision, which is important for experience goods. The objective of this paper is to rank online retailers based on apparel quality evaluative criteria. For this, a multi-step methodology is adopted. Initially, topic modeling through Latent Dirichlet Allocation (LDA) is applied on the accessed women apparel dataset for extracting evaluative criteria. LDA resulted in criteria namely aesthetic, functionality, expressive, performance, extrinsic, and return policy. Moreover, a problem that online customers face while making a purchasing decision is to choose the best retailer from the abundant options. Thus, in this study, BWM (Best Worst Method) technique is utilized for calculating criteria weights and R-TOPSIS (Technique for Order Preference by Similarity to Ideal Solution) is utilized for obtaining ranks of online apparel retailers. Further, the validation of the ranking method is done through CODAS (Combinative Distance based Assessment) method. Some implications for researchers and practitioners are provided at the end.