Defective products identification framework using online reviews

Identification of product defects has increasingly received researchers ' attention and assists various stakeholders in effectively removing product defects. This study proposed the robust defective products' identification framework and examined un-investigated review textual features. An innovative set of discrete emotions, psychological, and linguistic features are proposed. According to literature, these features were not explored for identification of defective products. An algorithm is derived to extract and compute the proposed features from the review text, and is tested using two examples. Moreover, a novel dataset of Amazon reviews is prepared by crawling reviews from four popular categories of products. Correlation and information gain statistical measures are used to select the subset of most influential features. The findings indicate that psychological indicators are more helpful than linguistic and discrete emotions as a stand-alone model in identifying product defects. In addition, the proposed indicators outperformed the state-of-the-art baseline. The baseline study used nine linguistic features for product defect identification. The findings reveal that the best five features are emotional tone, positive emotions, negation words, affective process and negative emotions. The implications of this research will help manufacturers, quality management and retailers to deliver their customers with defect-free products.