Type:

Year:

Supervisor:

Institution:

Page(s):

Abstract:

With the development of e-commerce, numerous business domains are looking for using at best the data generated by customers on the internet. Containing a large amount of information regarding user requirements and preference, online product review data are valuable for product designers. Comparing with the traditional user requirement identification methods like the focus group, questionnaire, and interview, these data have unprecedented characteristics: they are large in volume and they are renewing in real-time. The purpose of this study is to develop a design-oriented online review analysis approach to get useful insights based on the unprecedented characteristics of the online review data into product improvement and innovation. The proposed approach consists of two stages: data structuration and data analytics. The objective in the stage of data structuration is to mine and organize the words and expressions related to user requirements and preference from the unstructured review sentences. Only the structured data can be used for further analysis. In this research stage, an ontological model is firstly proposed to formalize the entities, properties and relationships of the words and expressions describing user requirements mentioned in the review sentences. The model consists of five concepts widely used through the process of design: product feature, product affordance, usage condition, user perception and user emotion. Then, a rule-based natural language processing method is proposed to identify automatically the words and expressions related to these five concepts. Experiments show that the performance of the proposed rulebased method is comparable to the previous studies. It provides designers with more information regarding user requirements to support decision-making. In the stage of data analytics, the author proposes two methods to process the structured data to obtain 1) users’ innovative usage of the product, which can inspire innovation path; 2) evolution of user preference on product affordances, which is useful for setting up product improvement strategies. The first method uses semantic similarity evaluation and classification algorithms to identify the product affordances that are mentioned less frequently. The second method innovatively applies traditional conjoint analysis to quantitatively categorize product affordances into the Kano model. Case studies with the online reviews of Kindle Paperwhite e-readers downloaded from amazon.com demonstrate the applicability of the two proposed methods in practice. Comparing with traditional user requirement identification methods, this study provides designers additional knowledge for decision making during product development based on the unprecedented characteristics of online review data. Industry can directly benefit from the design-oriented online review analysis approach proposed in this research project. The research trail may also serve as a guide for further research in the domain of design-oriented online review analysis.

Keywords:

online reviews, innovation, design engineering, natural language processing, opinion mining