Uncorrected Author Proof Journal of Intelligent & Fuzzy Systems xx (20xx) x–xx DOI:10.3233/JIFS-179352 IOS Press 1 Extracting user requirements from online reviews for product design: A supportive framework for designers 1 2 3 Kieu Que Anh a,∗ , Yukari Nagai b and Nguyen Le Minh b 4 a Japan Advanced Institute of Science and Technology, Japan 5 b Japan Advanced Institute of Science and Technology 6 Abstract. With the development of social networks and online shopping sites, we can easily obtain valuable feedback from users. The crucial question is how to utilize customer feedback for supporting the development of product design in the early phases. For product design, understanding user needs or user requirements would help designers design a better product for users. Therefore, user requirements is considered as an important role in product design. This paper proposes a framework for assessing user requirements from websites to support designers. They key idea is to extract user requirements from online customer reviews and represent them into an appropriate form for designers. We show that a support system consisting of feature aspect extraction, opinion summarization, and sentiment classification would be an useful tool for product design. Experimental results on a the data collected from the Amazon website show that supporting of opinion extraction techniques would be useful for designers in product design. 7 8 9 10 11 12 13 14 15 Keywords: Opinion mining, product design, sentiment classification, users requirements (URs) 16 1. Introduction 17 User requirements (URs) have been indicated that 18 it is an necessary component in the early product 19 design process [3, 29]). The main disadvantages of 20 traditional methods for capturing URs are that they 21 are time consuming, expensive, and not accessible 22 to large scale of opinions. For example, to obtain 23 valuable feedbacks for the product, we need to meet 24 users for interviewing them, therefore it is difficult 25 to collect a large scale of opinions. In addition, it 26 is very difficult to select suitable users from whom 27 to obtain their requirements. Fortunately, with the 28 development of the Internet, we can easily receive 29 valuable feedback from users across the world, by 30 processing review documents through a lot of online- 31 ∗ Corresponding author. Kieu Que Anh, Japan Advanced Insti- tute of Science and Technology, Japan. E-mail: queanhk@gmail. com. shopping sites such as Amazon.com and Ebay.com. 32 The problem, however, is how to extract URs from 33 raw texts to support designers? The initial work on 34 processing a products reviews mainly focuses on the 35 task named sentiment classification which aims at 36 classifying opinion documents into positive and neg- 37 ative. The output of this task may not be sufficient 38 for product designers to enrich their development. 39 There have been some works on extracting more 40 useful opinion texts to support users. An opinion min- 41 ing approach described in ([6]) showed that product 42 features were able to extract automatically. Another 43 work has recently indicated the importance of opin- 44 ion mining ([15, 23]) for business intelligence. The 45 work described in ([33]) showed that multiple opinion 46 text summarization can obtain useful text segments. 47 The authors conducted the experiment on the same 48 data described in [6] and gave a conclusion that their 49 approach appears to be more suitable than previous 50 ISSN 1064-1246/19/$35.00 © 2019 – IOS Press and the authors. All rights reserved