基于因果推断的多标记因果类属属性学习算法  

Multi-label causal generic attribute learning algorithm based on causal inference

在线阅读下载全文

作  者:鲍家朝 柏琪 BAO Jiachao;BAI Qi(School of Computer and Software Engineering,Anhui Institute of Information Technology,Wuhu 241100,China)

机构地区:[1]安徽信息工程学院计算机与软件工程学院,安徽芜湖241100

出  处:《黑龙江工程学院学报》2024年第5期16-24,共9页Journal of Heilongjiang Institute of Technology

基  金:安徽省自然科学基金项目(2022AH051894);安徽省高校自然科学研究项目(KJ2021A1207)。

摘  要:真实世界中的事件间存在着因果关系,特征与标记间也存在着潜在的因果关系。基于此,提出一种基于因果推断的类属属性学习算法CLSF。类属属性算法充分挖掘各标记的独有特征并构建分类器,从而在一定程度上提升多标记学习的性能。传统的类属属性算法大多依赖于欧氏距离或L1范数提取类属属性。然而,欧氏距离并不适用于高维空间的度量。而L1范数的稀疏性很大程度上取决于参数。因此,文中算法使用因果推断学习类属属性。首先对原始标记进行增强化处理,并基于增强标记学习标记和特征之间的因果关系,然后通过因果关系矩阵约束权重矩阵提取因果类属属性。多个多标记基准数据集中的实验表明,CLSF算法较其他对比算法具有一定的性能优势,统计假设检验的结果也表明文中算法的有效性。Similar to causal relationships that commonly exist in the world,causal effect also exists between features and labels.Based on this,a causal inference-based label-specific features learning algorithm CLSF(Causal Label-Specific Features Learning)is proposed in this paper.Label-specific features learning algorithm fully explores the unique characteristics of each label,classifiers trained with label-specific features usually gain an improved performance to a certain extent.Euclidean distance and L1-Norm are the most common methods for existing label-specific features learning algorithms.However,Euclidean distance is insufficient for measuring high-dimension data.Meanwhile,L1-Norm requires a proper parameter election.Therefore,in this paper,we utilize the causal inference method to learn label-specific features.First,we embed the original label into a new space to acquire continuous labels.Then the causal relationship between features and labels is obtained with a causal inference method.Finally,it constrains the original feature space with the learned causality to extract causal label-specific features.Experiments conducted on multiple multi-label benchmark datasets demonstrated that the CLSF algorithm has some advantages over other comparison algorithms.The statistical hypothesis testing results also further prove the proposed method’s effectiveness.

关 键 词:类属属性 标记增强 因果推断 多标记学习 

分 类 号:TP181[自动化与计算机技术—控制理论与控制工程]

 

参考文献:

正在载入数据...

 

二级参考文献:

正在载入数据...

 

耦合文献:

正在载入数据...

 

引证文献:

正在载入数据...

 

二级引证文献:

正在载入数据...

 

同被引文献:

正在载入数据...

 

相关期刊文献:

正在载入数据...

相关的主题
相关的作者对象
相关的机构对象