ABSTRACT Kansei Engineering (KE) is a user-oriented technology combing customer psychological feelings and engineering for designing and developing products. Conventionally, questionnaire surveys have been extensively applied for understanding customers’ affective demands, responses and evaluations. However, the questionnaire is usually time-consuming, labour-intensive and small in data size. Online customer reviews provide trustable, continuously updated and free customers’ responses. Existing studies generally focus on the polarity classification of the positivity and negativity of the review texts. This study proposes an opinion mining approach based on KE and machine learning to extract and measure users’ affective responses to products from online customer reviews. Five types of machine learning algorithms are applied, including Support Vector Machine (SVM), Support Vector Regression (SVR), Classification and Regression Tree (CART), Multi-Layer Perceptron (MLP) and Ridge Regression (RR). An experiment has been conducted to illustrate the proposed approach. The results show that SVM+SVR is the best performer. It achieved a recall, precision and score of more than 80% for the classification of the soft-hard attribute with the smallest mean square error. Based on the proposed method, designers and manufacturers can effectively know customers’ responses to products through inputting the review texts to facilitate the process of product design.

中文翻译：

---

从客户评论中提取情感反应：意见挖掘和机器学习方法

摘要 Kansei Engineering (KE) 是一种以用户为导向的技术，将客户心理感受与工程学相结合，用于设计和开发产品。传统上，问卷调查已被广泛应用于了解客户的情感需求、反应和评价。然而，问卷通常耗时、劳动密集且数据量小。在线客户评论可提供值得信赖、不断更新和免费的客户回复。现有的研究一般集中在评论文本的积极性和消极性的极性分类上。本研究提出了一种基于 KE 和机器学习的意见挖掘方法，从在线客户评论中提取和衡量用户对产品的情感反应。应用了五种类型的机器学习算法，包括支持向量机 (SVM)、支持向量回归 (SVR)、分类回归树 (CART)、多层感知器 (MLP) 和岭回归 (RR)。已经进行了一个实验来说明所提出的方法。结果表明，SVM+SVR 的表现最好。对于均方误差最小的软硬属性分类，实现了80%以上的召回率、准确率和分数。基于所提出的方法，设计师和制造商可以通过输入评论文本来有效地了解客户对产品的反应，从而促进产品设计过程。结果表明，SVM+SVR 的表现最好。对于均方误差最小的软硬属性分类，实现了80%以上的召回率、准确率和分数。基于所提出的方法，设计师和制造商可以通过输入评论文本来有效地了解客户对产品的反应，从而促进产品设计过程。结果表明，SVM+SVR 的表现最好。对于均方误差最小的软硬属性分类，实现了80%以上的召回率、准确率和分数。基于所提出的方法，设计师和制造商可以通过输入评论文本来有效地了解客户对产品的反应，从而促进产品设计过程。