基于余弦相似度的因子分析在食品成分检测中的应用

Dynamical Clustering by Cosine Similarity Algorithm

余弦相似度聚类不但体现了向量之间的相似关系,而且包含了向量内部元素的变化状况。研究了以角距离余弦为相似度的动态聚类方法,探讨了该方法的原理,研究了该方法的具体算法步骤,并利用大米样品的近红外光谱数据进行了验证。结果表明,通过对49个大米样本的光谱数据进行余弦相似度聚类分析,得到了9个特征因子,经与大米蛋白质含量参比值回归,其标准差为0.3,平均相对误差为2.3%,相关系数为0.9548。

Cosine similarity is a good reflection of vector's similarity, and it also reflect the variations of vector's elements. Using cosine similarity as a scale of clustering could better tell how many variations contained in the vectors and could find some eigenvectors to substitute all vectors in evaluation. The eigenvectors found by clustering could be used in regression. This article described a means of dynamically clustering with Cosine Similarity, forty-nine rice samples' infrared spectra had been used to test the means of clustering, and 9 independent clusters were identified.