generated by a function of the distances along some set of dimensions, then the resulting configuration of points in a scaling solution should reflect those dimensions. However, often Euclidian distances are used in scaling procedures so that the orientation of axes in the resulting configuration is arbitrary: any rotation of the axes would result in the same distances (and therefore stress). In such cases, the researcher can either visually scan the configuration in order to choose an orientation of axes that leads to interpretable results or apply less arbritrary procedures by multiple regression analyses. In such analyses, the idea is to regress meaningful variables on the coordinates for the different dimensions and rotate the solution as to maximize the interpretability. In Figure 1c, the two dimensional scaling solution is shown for the 10 faces. After visual inspection, the configuration can be interpreted as the perceptual dimensions of age and adiposity.
become relative to N in order for this technique to make drastic savings possible in the similarity data collection.
Another recent advance in MDS models is the feature mapping approach (Rumelhart & Todd, 1992; Steyvers, & Busey, 2000). In the traditional approach, the physical representation of the features comprising the stimuli is explicitly ignored. In such a purely top-down approach, the multidimensional representations are sometimes difficult to relate back to the physical stimulus by visual interpretation or regression analyses. In the feature mapping approach, in addition to the proximity data, additional physical measurements on the set of stimuli is available. The goal is to find a mapping between the set of physical measurements of an stimulus to the position of that stimulus in an abstract psychological space. With this approach, the multidimensional space is related directly to the physical dimensions of the stimuli.
Advances in MDS
The success of the MDS approach arises in part from the simplicity of the underlying assumptions and the wide availability of computer software to create scaling solutions. Recent research has expanded the scope of the MDS approach in several directions. In Isomap (Tenenbaum, De Silva, & Lanford, 2000), stimuli are represented as points lying on a non-linear manifold in some multidimensional space. Similarity then is computed as the geodesic distance on the manifold (i.e., the shortest distance along the manifold) as opposed to Euclidian distance in MDS. This technique is capable of discovering the non-linear degrees of freedom that underlie complex data-sets.
A drawback of most MDS algorithm is that a N x N matrix of similarity judgments is needed to scale N objects. Therefore, the number of similarity ratings needed depends quadratically on the number of objects which leads to practical limitations (e.g., subject time, number of subjects) on the number of objects that can be used in scaling studies. With modified MDS procedures, the amount of data that needs to be collected might be reduced. In the anchor point method (e.g., Buja, Swayne, Littman, & Dean, 1998), subjects rate all similarity pairs involving N objects and a smaller number of K anchor points that provide a representative sample of N objects. A modified MDS procedure then analyses the N x K similarity matrix in order to scale N objects. Future research will have to show how small K can
Challenges for MDS
Tversky and Hutchinson (1986) have argued that for language-related stimuli that have conceptual as opposed to perceptual relations, geometric models based on MDS may fail to capture some aspects of the data and might therefore be inappropriate as a representational basis. For such stimuli, tree or graph-theoretic structures might be better suited than spatial/dimensional models based on MDS. Nevertheless, the geometric model of similarity has been applied to a wide variety of stimuli rich in conceptual structure. In Latent Semantic Analysis (Landauer & Dumais, 1997), words are placed in a high dimensional semantic space by a procedure related to MDS by analysing the co-occurrence statistics for words appearing in contexts in a large corpus. In this semantic space, words with similar meaning are placed in nearby regions of the space. While the Latent Semantic Analysis approach has been very successful in modelling human performance in a variety of semantic tasks, it remains to be seen to what degree a geometric model is appropriate to model language related processes.
Also, for some highly structured objects, the simple geometric model for similarity as used in MDS may not be particularly revealing for analysing the process of generating similarity judgments. In a geometric model, it is assumed that the set of objects can be described by a fixed collection of feature values where the process of generating similarity judgments is
多维标度法(MDS)的改进及其在国际市场研究中的应用多维标度法(MDS)的改进及其在国际市场研究中的应用隐藏>> ? 1995-2004 Tsinghua Tongfang Optical Disc Co.,...
本章仅介绍常用的古典多维标度法和权重多维 标度法。 第二节 古典多维标度法 (Classical MDS) 一 相似与距离的概念 二 古典多维标度分析的思想及方法三 度量...
23页 免费 MDS多维标度 5页 免费 市场分析方法LV11 多维尺... 暂无评价 9页 免费 多维尺度分析 14页 1下载券喜欢此文档的还喜欢...
第十章 多维标度法第一节 第二节 第三节 第四节 引言 古典多维标度法(Classical 古典多维标度法 MDS) 权重多维标度(WMDS) 权重多维标度 实例分析与计算实现...
多维标度_经济/市场_经管营销_专业资料。第十章第一节 引言 多维标度法 第二节 第三节第四节 古典多维标度法(Classical MDS) 权重多维标度(WMDS) 实例分析...
本章仅介绍常用的古典多维标度法和权重多维 标度法。 第二节 古典多维标度法 (Classical MDS) 一 相似与距离的概念 二 古典多维标度分析的思想及方法三 度量...
多维标度法汇报学习_教育学/心理学_人文社科_专业资料。多维标度法 ? 第一节 第二节 第三节 引言 古典多维标度法(Classical MDS) 权重多维标度(WMDS) ? ...
第二章 多维标度法简介多维标度法(multidimensional scaling MDS)是一种把高维降为低维,在低维空间展 示距离数据结构的多元数据分析技术,简称 MDS。多维标度法...
多维标度法的聚类分析_问题与解法_工学_高等教育_教育专区 暂无评价0人阅读0次下载举报文档 多维标度法的聚类分析_问题与解法_工学_高等教育_教育专区。MDS ...
满足非度量多 维标度法对相异性数据的要求。因此本文提出的RSSICMDS 算法, 直接使用 RSSI 值进行定位。 基于 RSSI 构建多维标度法中的相异性矩阵 P , 计算出...
热门文档