Article Preview
TopIntroduction
Products such as automobiles may have both safety and performance defects. Government regulations and exposure to severe brand-value and financial losses compel manufacturers to be pro-active in detecting and eradicating safety defects. Traditionally, safety defects are identified through process improvement tools and service center feedbacks. Such approaches not only suffer from high cost, and incomprehensiveness; their applicability is limited in the case of performance defects (Law et al., 2017; Liu et al., 2018). In this regard, massive product review data generated from the web has turned out to be an important source to comprehend user experiences, reactions, and perceptions. While prospective consumers use them to analyse the peers’ experience with the product, the organizations mine it to identify user requirements and expectations (Singh et al., 2020). However, it is beyond human cognition to scan the available reviews manually, summarize them, and use them for sensible decision making.
In this regard, artificial intelligence in general (Dwivedi et al., 2019; Grover et al., 2019; Dwivedi et al., 2020Stieglitz et al., 2020) and sentiment analysis (SA) in particular has emerged as a tool to mine information from text. Its usefulness is well tested and validated in domains such as product promotions and marketing (Ting et al., 2014), demand and sales forecasting (Archak et al., 2011; Chong et al., 2017; Geva et al., 2013; Hou et al., 2017; Zhang et al., 2020), supply-chain performance evaluation (Swain & Cao, 2019), and product quality assessment (Abrahams et al., 2015; Law et al., 2017). Specifically, it assists the businesses in decision making in automotive industry (Abrahams et al., 2012, 2013, 2015; Gruss et al., 2018) electronic products (Abrahams et al., 2015), dishwasher appliances (Law et al., 2017), body wash products (Zhang et al., 2012), entertainment industry (Chintagunta et al., 2010; Yang & Chao, 2015) travel industry (Chang & Chen, 2019; Choi & Lee, 2017; Sann, & Lai, 2020), and the toy industry (Winkler et al., 2016; Saumya et al., 2019). However, there has been almost no effort to connect these results with traditional quality-control tools with which the manufacturing community is acquainted. Moreover, most of such studies focus on document or sentence level. More recently, aspect-level sentiment analysis (ASLSA) has emerged as a tool to identify product defects, more precisely targeting specific attributes and context (Schouten & Frasincar, 2016). In this research, the authors have contributed to this growing field by proposing an integrated automobile-defect detection framework that connects ASLSA with the Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS) and traditional quality-control tools. The framework answers following research questions:
RQ1: What are the important product features, which customers frequently discuss in online reviews?
RQ2: How feature level consumer sentiments be used to quantify manufacturers’ perceived performance rating?
RQ2: Are review embedded consumer sentiments useful in discovering products’ perceived weakness?