An integrated probabilistic graphic model and FMEA approach to identify product defects from social media data

摘要

Recently, the explosive increase in social media data enables manufacturers to collect product defect information promptly. Extant literature gathers defect information like defective components or defect symptoms without distinguishing defect-related (DR) texts from defect-unrelated (DUR) texts and thus makes defects discussed by few texts buried in enormous DUR texts. Moreover, existing studies do not consider the defect severity which is valuable and important for manufacturers to make remedial decisions. To bridge these research gaps, we propose a novel approach that integrates the probabilistic graphic model named Product Defect Identification and Analysis Model (PDIAM) with Failure Mode and Effect Analysis (FMEA) to derive product defect information from social media data. Comparing to extant studies, PDIAM identifies DR texts and then extracts defect information from these texts. And PDIAM provides more defect information than previous researches. Besides, we further analyze defect severity with the combination of FMEA and PDIAM which alleviates the inherent subjectivity brought by expert evaluation in the traditional FMEA. A case study in the automobile industry proves the predominant performance of our approach and great potential in defect management.