【作者】 石夫乾；

【导师】 孙守迁；

【作者基本信息】 浙江大学 ， 数字化艺术与设计， 2011， 博士

【摘要】 传统感性工学的研究是利用成对感性词汇(Paired KANSEI Adjectives, PKA)来建立调查表,以期获得用户对产品的感性评价。但是,这种方法仅是利用感性词汇集对事物在感性语义上的一个粗粒度的定性描述,所利用的尺度是离散且线性的；这种方式的感性知识建模存在着极大的不精确与不确定性。本文将从基础数据获取、产品内隐情感表示方法及推理模型的建立三个方面入手建立一个全新的产品内隐情感认知模型。首先,利用情绪词库(Affective Norms for English Words, ANEW)所采用的实验方法,开展心理量的人工标定实验；建立起人体在“自然状态”下,直观感知(Valence-Arousal, VA,即Valence表示兴奋或平静的程度, Arousal表示与正面或者负面的程度)与感性词汇之间的联系。其次,利用心理量人工标定实验所获得的Valence-Arousal数据,建立起基于Valence-Arousal的情感细胞元认知模型,包括基于情感细胞元模型的内核和外壳的定义及其度量函数的学习方法,同时对于细胞元的边界模糊性(Fuzziness)进行计算；建立起基于Valence -Arousal情感细胞元的产品内隐情感表示方法；在情感细胞元模型之上,给出了基于模糊集理论(Fuzzy Sets, FS)的相似关系,并通过相似性计算与分析给出了产品内隐情感检索系统设计方法。最后,利用案例式学习技术(Case Based Reasoning, CBR)、贝叶斯变精度粗糙集合成方法及模糊多分类支持向量机(Multi-class Fuzzy Support Vector Machine, MFSVM)技术,分别建立起基于Valence-Arousal的产品内隐知识推理模型,并在手机、汽车等产品的案例研究中验证其有效性。本文的研究成果包括：(1)建立了基于Valence-Arousal的情感细胞元的产品内隐情感表示模型。这种模型是对“离散”的感性词汇的深入解析,是意图在定性与定量之间寻求一个平衡点,使得每一个感性词汇不再是孤立的、生硬的,而是一个“细胞”。(2)建立起基于Valence-Arousal的情感细胞元的内核与外壳的获取方法。包括建立了Valence-Arousal二维情感函数到情感细胞元的映射关系。针对内核的单点集、平面集和圆型集的不同类型分别进行定义；通过单一和混合高斯密度函数来度量情感空间点集的密度,并通过参数估计的方法来获得密度函数的关键参数。(3)建立起情感细胞元边界的模糊计算模型。通过模糊集的模糊交、并等逻辑运算建立情感细胞元模糊边界的计算模型。针对单点集、平面集和圆型集分别给出边界的计算模型。情感细胞元的边界计算对于产品内隐情感的精确分类起到重要作用。(4)建立基于Valence-Arousal的情感细胞元相似性度量方法。包括相似性度量空间、相似关系的形式化定义和相似度的计算。在IF-THEN规则的知识表达形式中,对相似性计算的结果进行分析,并给出产品内隐情感检索系统的研究内容与设计方法。(5)建立起三种基于Valence- Arousal的产品内隐情感推理模型。在模糊案例式推理模型中,以产品造型典型特征集为初始案例,以相应地情感细胞元密度函数生成案例相似度计算公式,通过模糊最邻近算法(Near Neighbor, NN)算法获得产品的整体评价。在贝叶斯学习模型中,通过基于情感细胞元的产生式规则及其度量(LN,LS)并利用可变精度粗糙集理论形成贝叶斯变精度合成推理模型。在多分类器推理技术中,将细胞元的密度分布函数融入多分类模糊支持向量机模型,建立了产品内隐情感的分类模型。更多还原

【Abstract】 The objective of traditional KANSEI engineering is to create a questionnaire to obtain users’ perceptual evaluation of products by using Paired Kansei Adjectives (PKA). However, this approach is only to use the emotional vocabulary of things under a coarse-grained description in the emotional semantic characterization, the use of the scale is discrete and linear; perceptual knowledge modeling in this way exits significant imprecision and uncertainty.This thesis proposed a new cognitive model of emotional semantics on three aspects:the basic data acquisition, representation and reasoning model of product implicit emotion.First, the emotional thesaurus (Affective Norms for English Words, ANEW) experimental methods was adopted to conduct psychological experiments by artificial calibration and to establish the connections between body’s visual perception (Valence-Arousal, which valence denotes the degree of excited or calm, arousal denotes the degree of positive or negative) under the "natural state" and emotional words. Second, the emotional Valence-Arousal based emotional cellular model was established by using Valence-Arousal data which obtained from the manual calibration psychological experiments, including emotional cellular-based kernel and shell definition and its learning methods of measurement functions; at the same time, the boundaries of the ambiguity of a cellular (Fuzziness) was calculated. Third, the representation model of implicit emotion was founded based on Valence-Arousal emotional cellular; a fuzzy similarity analysis of emotional cellular based on fuzzy set was introduced and the design methodology of an implicit emotional retrieval system of product was proposed. Finally, reasoning model of products implicit emotion based on Valence-Arousal was presented respectively by using CBR (Case Based Reasoning), Bayesian synthesis based variable precision Rough Sets method and MFSVM (Multi-classifier Fuzzy Support Vector Machine) techniques are given to demonstrate the effectiveness of the proposed methodology in cases study of mobile phone and vehicle design.The main contributions of this thesis include:(1) Established the Valence-Arousal based emotion cellular representation model of product implicit emotion which is the intention and in-depth analyzing of these "discrete" KANSEI adjectives that is to find a balance point under the qualitative and quantitative analysis, so that each word is no longer isolated emotional, blunt, but a "cell."(2) Established the kernel and shell’s definition and its acquisition method of Valence-Arousal based emotion cellular, which included the mapping relations between emotional cellular and the two-dimensional Valence-Arousal emotional space and the definitions of three type kernels (single point type, flat type and circle type). The Gaussian singular and mixture models were applied to describe the density of points on Valence-Arousal space for difference types and the parameters of Gaussian singular and mixture model are acquired by parameters estimation methods.(3) Proposed the computing methods of fuzzy boundary of emotional cellular model, which included establishing boundaries and the calculation model by fuzzy set and fuzzy logic operations such as intersection, union; and the calculation process of boundary for the kernels on single-point type, flat type and circle type respectively. It plays an important role for implicit emotion computing to acquire the exact classification of emotions by using boundaries computing of emotional cellular.(4) Established similarity measure model between Valence-Arousal based emotional cellulars which included the definition of similarity metric space, the formal definition of similarity relation and similarity calculations; introduced a similarity computing model on IF-THEN rules of knowledge representation and also demonstrated the design methodology of the product’s implicit emotional retrieval system by using the proposed similarity computing model.(5) Established three kinds of reasoning models on product’s implicit emotion retrieving based on Valence-Arousal. In the fuzzy case based reasoning model, a product typical feature set was regarded as the initial case, the corresponding density function of emotion cellular was applied to calculate the similarity degree and obtain the overall evaluation of product by using the fuzzy nearest neighbor algorithm (Near Neighbor, NN). In the Bayesian learning model, rule based representation of production emotion cellular and its measurement (LN, LS) employing variable precision rough set theory was used to this synthesis model. As in multi-classifier model, the multi-classifier fuzzy support vector machine technology combined density distribution function of emotional cellular was applied to build the classification model of product’s implicit emotion.更多还原