Back to results list
Show full item record
Please use this identifier to cite or link to this item:
|Title:||Information mining from online reviews for product design||Authors:||Jin, Jian||Degree:||Ph.D.||Issue Date:||2013||Abstract:||To improve product quality, designers should understand customers' needs. Traditionally, customers' needs are gathered through surveys. Nowadays, more and more e-commerce websites encourage customers to post online reviews and express their opinions. Consequently, online reviews become a valuable source of customers' needs. However, it is difficult for product designers to understand all the relevant customers' needs from a vast number of online reviews. Under this circumstance, an intelligent system should be designed to mine useful information from online reviews and help designers to improve their products. To understand how designers filter and digest the customers' needs, two exploratory case studies are conducted. The purpose of the first case study is to explore why some reviews are preferred by product designers, while the second is to understand how designers analyze online reviews. From the first case study, two questions for the identification of design-preferred reviews are clarified. (1) The first question is how to identify helpful online reviews from designers’ perspective. This question is formulated as a regression model. According to designers’ arguments about why some reviews are helpful, the regression model is built from four categories of features which are extracted directly from the review contents. Closely associated with this question, another concern is whether the concept of helpfulness perceived by designers in one domain can be migrated to other domains. Different methods of feature selection are employed for this concern. (2) The second question is how to recommend rating values on online reviews by taking designers’ personal preferences into consideration. This question is formulated as a classification model. Utilizing the four categories of features, this classification model is considered from both the generic helpfulness aspect and the personal preference aspect from a designer. Various experiments suggest that design-preferred reviews can be identified by analyzing review content automatically.
From the second case study, two questions for building a design-centered knowledge base from online reviews are explored. (1) The first question is how to associate online reviews with product characteristics. A probabilistic approach, which utilizes the statistic information about keywords and context words in the online reviews, is proposed for this question. The impacts of context words are estimated in this probabilistic approach according to their distances to keywords. (2) The second question is how to prioritize product characteristics from online reviews. Based on the customer satisfaction on online reviews, an ordinal pairwise supervised classification approach is developed for this question. Also, an integer nonlinear programming optimization model is advised to make the pairwise-based results of this approach to be evaluated with standard classification and ranking evaluation metrics. The encouraging results validate the feasibility of the proposed methods. Overall, in this research, a regression model is proposed to identify helpful online reviews; several feature selection methods are compared to explore whether the concept of helpfulness perceived by designers in one domain can be migrated to other domains; a classification model is suggested to recommend rating value on online reviews; a probabilistic approach is developed to connect online reviews with product characteristics; and finally, an ordinal pairwise supervised classification approach as well as the integer nonlinear programming optimization model is advised to prioritize product characteristics based on online reviews. The proposed techniques benefit product designers to improve their products and attract more customers. Future extensions can be conducted towards building intelligent applications to process online reviews for product designers.
Hong Kong Polytechnic University -- Dissertations
|Pages:||xviii, 299 leaves : ill. ; 30 cm.|
|Appears in Collections:||Thesis|
View full-text via https://theses.lib.polyu.edu.hk/handle/200/7198
Citations as of May 15, 2022
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.