Using unlabeled data mining to detect customer perceptions of undefined commodity problems

Abstract

Understanding how a customer perceives an undefined commodity problem is important for online retailers so that they can address problems and satisfy and gain customers. Data mining technological maturation and developments in online review systems means that it is now possible to mine for customer perceptions on commodity problems from structured and unstructured data. This research, therefore, mainly used an unsupervised machine learning, stacked denoising autoencoder-K-means, to resolve the customer perception process for undefined commodity problems. It was found that: 1) textual reviews and quantitative scores are mutually complementary when analysing online buyer perceptions; 2) customer perception systems have a typical line-of-sight to capture the undefined commodity problem attributions. Although the attributions related to undefined commodity problems are very scattered, a highly unified strategy, providing after-sales service, was found to exist within each group, which was agreed through group consensus by about 98% of the consumers.