Data-driven methodologies for decision making in engineering design

In the product development process, customer needs are essential to develop the product concepts. These concepts are crucial because the subsequent stages in the process are dependent on the selected concepts. Customer needs are conventionally gathered via survey-based methods, which may require extensive cost to conduct. Along with the massive growth of internet, an alternative to those survey-based methods emerges. Customer needs, as well as other insights about the customers, may be inferred from the opinions, feedbacks, or expectations that customers express in various online channels including online customer reviews. However, the volume and the generating velocity of the online customer review data surpass people's ability to analyze them in a reasonable time. Therefore, in order to utilize online reviews for supporting product designers in decision making, this work proposes methodologies that utilize Natural Language Processing tools, machine learning algorithms, and statistical models. In particular, the methodologies are proposed to support product designers in three specific aspects. First, a methodology is proposed to automatically identify product features that are discussed in the customer reviews as well as their corresponding sentiments. The particular product features that are significantly related to sales rank should become the focus of product designers when considering improvements of the existing product. Second, a novel approach to constructing the choice sets in the absence of both socio-demographic and the actual choice set data is proposed. The choice models that use the proposed choice sets are shown to have better predictive ability than the baseline, i.e., using random choice sets. The choice models with higher predictive ability are useful for product designers to perform demand estimation more accurately. Finally, a methodology is proposed to automatically identify product usage contexts from online customer reviews. Understanding the actual usage contexts is important because it may explain the differences in customer needs, the required design targets, and product preferences. In this work, the identified usage contexts are further complemented by their corresponding aspect sentiments. For product designers, the results enable them to understand customer experience regarding the usage contexts, including the contexts that may not be originally intended by the designers