Abstract
This paper describes a method to use data-mined customer reviews to identify potential product defects. The process involves locating negative reviews of a specific product and then extracting comments that have potential connections with the product design. The extracted comments are then categorized and correlated with features of the product. Given that the customer comments are generally not specifically tied to engineering requirements, or even posed in engineering terms, this correlation requires some degree of engineering analysis to establish a correlation. After these complaint-feature correlations are established, then engineering tests directed at understanding the possible defects are performed. This data-mining process effectively harnesses the massive amount of in-situ testing and evaluation that is performed by customers of the product. The process is illustrated via a case study of an infant bassinet. Customer reviews of the product were studied to identify and categorize key complaints about the bassinet. These complaints were correlated with features and potential defects in the product, including assembly difficulties, as well as a sleeping surface that tilts and causes infants to roll and press into the mesh side wall while sleeping. Then, assembly and sleeping surface deflection tests were conducted. The sleeping surface deflection tests include measurements of how bassinet leg separation affects the deflection, as well as long-duration progressive deflection that occurs from repeated use of the product. The results of the engineering testing confirm the presence of defects in the bassinet and its assembly instructions, as suggested by the reviews. This case study illustrates how data-mined customer reviews provide a valuable source of engineering data and indications of product defects.