Product defects are a major concern for manufacturers and customers. Detecting product defects is vital for manufacturers to prevent enormous product failure costs. As the surge of social media is in vogue, social media data become an important information source for manufacturers to collect defect information. In this study, we propose a novel probabilistic graphic model to discover defects from social media data. We first use three filters, namely, sentiment filter, component-symptom filter and similarity filter, to select informative data. Second, we analyze the remaining data via the proposed probabilistic graphic model and identify defect-related data. Our method provides detailed defect information including defect types, defective components and defect symptoms which is omitted by previous research. A case study in the automobile industry validates the effectiveness and superior performance of our method compared to prior approaches.

产品缺陷是制造商和客户的主要关注点。检测产品缺陷对于制造商防止巨大的产品故障成本至关重要。随着社交媒体激增，社交媒体数据成为制造商收集缺陷信息的重要信息来源。在这项研究中，我们提出了一种新颖的概率图形模型，以从社交媒体数据中发现缺陷。我们首先使用情感过滤器，分量症状过滤器和相似性过滤器这三个过滤器来选择信息数据。其次，我们通过提出的概率图形模型分析剩余数据，并确定与缺陷相关的数据。我们的方法提供了详细的缺陷信息，包括缺陷类型，缺陷组件和缺陷症状，而先前的研究已将其省略。