多维标度法在可视化卷烟感官评价结果中的应用

Abstract

卷烟感官评价是一种多指标评价，以便较全面表征卷烟的各方面感官特征。但由于评价结果为多维，不易于直接比较多种不同卷烟之间的不相似度。为此，对全国54个代表性卷烟样品的评价结果，比较研究了多种多元统计作图技术的表现。结果表明：1) 统计脸谱图、星形图、平行坐标图等无损作图法仅适合少数样品的可视化比较；2) 无损作图法中，采用主成分分析法时，前几个主成分解释的方差比例比较小，需要至少在十维空间才能保持原始数据的信息不会有严重损失。因此，采用主成分分析在二维空间中进行可视化会严重失真；采用多维标度法在四维空间就能较好保留原始空间中两两样品的不相似度，因此，多维标度法更适合对卷烟感官评价结果进行可视化。 Cigarette sensory evaluation is a multi index evaluation, so that the various aspects of cigarette sensory can be characterized comprehensively. However, the evaluation results are multi-dimen- sional, and it is not easy to compare the differences of different cigarettes directly. For the national evaluation results of 54 representative samples of cigarettes, there is a comparative study of var-ious multivariate statistical mapping techniques. The results showed that: 1) Some lossless graphing methods such as Facebook statistics chart, star chart, parallel coordinate plots so on are only suitable for visual comparison in a small number of samples; 2) In the nondestructive mapping method, when using principal component analysis, the variance proportion of the first few principal components explained is relatively small. It means that we need at least ten dimensions space, in order to keep the original data as more information as possible. Therefore, the use of principal component analysis to visualize severely distorted in two-dimensional space; using mul-tidimensional scaling method will be able to better preserve the original space dissimilarity between two samples in four-dimensional space. In conclusion, multidimensional scaling method is more suitable for visualization of cigarette sensory evaluation results.