Bi-channel dynamic topic model for quality monitoring considering initial and additional online customer reviews

E-commerce provides a vast space for user-generated content, including user reviews from the initial use of products or services to subsequent usage. Existing methods mainly focus on unmodified reviews and ignore customer perception after more experience with products. To address this, we propose a Markov chain-based bi-channel dynamic topic model (BDTM), which extends the sequential structure of the dynamic topic model and incorporates initial customer reviews and additional reviews to reflect topic shifts caused by customer experience perception. Then, a control chart based on word mover's distance (WMD), called a BDTM-WMD (B-W) chart, is proposed to monitor topic shifts under BDTM. An alternative multiscale dynamic topic model (M-K) chart is constructed for comparison. Using a simulation approach, we find that compared with the existing sequential reverse joint sentiment-topic (SRJST) chart and joint sentiment topic-rating meets review (JSTRMR) chart, the proposed B-W chart is more sensitive to small shifts. Case study 1, using real data, shows the advantages of the proposed BDTM, B-W chart and M-K chart under both initial and additional reviews. Case study 2 shows that with additional reviews, the proposed B-W chart triggers out-of-control signals earlier than those by the existing joint sentiment topic-rating meets review (JSTRMR) chart.