Detecting and prioritizing product defects using social media data and the two-phased QFD method

Abstract

Recently, social media data have been leveraged for product defect discovery. But the considerable number of defects reflected via social media inhibits manufacturers from solving product defects promptly. Extant studies focus on identifying defect-relevant texts and then deriving defects discussed in texts. They omit to assess the importance of discovered defects and find the defects with high priorities. In this study, we first developed a topic model named Defect Analysis Model (DAM) to discover product defects from defect-related texts, which are identified by the integrated BERT and Random Forest. Then we propose the Two-Phased Quality Function Deployment for Defect (TPQFDD) to prioritize discovered product defects. With the consideration of defect frequencies and defect costs, TPQFDD evaluates the importance of defective components, defects, and defect causes for more inspired managerial insights. We compare our approaches with baseline approaches using an online thread dataset of automobiles. Comparison results prove that our methods effectively detect and prioritize product defects that occurred in the aftersales stage.