摘要
为解决传统非负矩阵分解不考虑潜在因子的相关性与分布特征等缺点,提出一种基于最大熵与相关性分析的非负矩阵分解方法。利用最大熵原理描述非负矩阵分解中的潜在因子分布,以捕捉语义质量的潜在因子特性,并提出一种基于相似性的方法来度量差异性。将自适应加权策略引入因子间的相互关系,使得每个潜在因子能够无监督地获得自适应权重,并对自适应加权的潜在因子进行非线性变换。在多个数据集上的实验结果表明,该方法能够提升传统方法的效果。
In order to solve the shortcomings of traditional non negative matrix decomposition,such as ignoring the correlation and distribution characteristics of latent factors,a non negative matrix decomposition method based on maximum entropy and correlation analysis is proposed.The distribution of latent factors in non negative matrix factorization was described by maximum entropy principle to capture the latent factors characteristics of semantic quality.A similarity-based method was proposed to measure the difference.The adaptive weighting strategy was introduced into the relationship between factors,so that each latent factors could obtain the adaptive weight unsupervised.The latent factors of adaptive weighting were transformed to nonlinearity.A large number of experimental results on multiple datasets show that the proposed method can improve the effect of traditional methods.
作者
冯本勇
徐勇军
Feng Benyong;Xu Yongjun(Shijiazhuang Vocational College of Industry and Commerce,Shijiazhuang 050000,Hebei,China;Institute of Computing Technology,Chinese Academy of Sciences,Beijing 100190,China)
出处
《计算机应用与软件》
北大核心
2023年第9期267-277,共11页
Computer Applications and Software
基金
国家自然科学基金项目(61702487)。
关键词
非负矩阵分解
最大熵原理
自适应加权
潜在因子
相关性
Nonnegative matrix factorization
Maximum entropy principle
Adaptive weighting
Latent factors
Correlation
