特征标记依赖自编码器的多标记特征提取方法

在多标记学习中,如何处理高维特征一直是研究难点之一,而特征提取算法可以有效解决数据特征高维性导致的分类性能降低问题。但目前已有的多标记特征提取算法很少充分利用特征信息并充分提取"特征-标记"独立信息及融合信息。基于此,提出一种基于特征标记依赖自编码器的多标记特征提取方法。使用核极限学习机自编码器将原标记空间与原特征空间融合并产生重构后的新特征空间。一方面最大化希尔伯特-施密特范数以充分利用标记信息;另一方面通过主成分分析来降低特征提取过程中的信息损失,结合二者并分别提取"特征-特征"和"特征-标记"信息。通过在Yahoo多组高维多标记数据集上的对比实验表明,该算法的性能优于当前五种主要的多标记特征提取方法,验证了所提算法的有效性。

Multi-Label Feature Extraction Method Relied on Feature-Label Dependence Autoencoder

In multi-label learning,how to deal with high-dimensional features has always been one of the research difficulties.The feature extraction algorithm can effectively solve the problem of classification performance degradation caused by high dimensionality of data features.However,the existing multi-label feature extraction algorithms rarely make full use of feature information and fully extract the“feature-label”independent information and fusion information.Based on this,a multi-label feature extraction method based on feature-label dependence autoencoder is proposed.The kernel extreme learning machine self-encoder is used to fuse the label space with the original feature space and generate the reconstructed feature space.On the one hand,Hilbert-Schmidt independence criterion is maximized to make full use of the information between labels and the features;on the other hand,principal component analysis is used to reduce the information loss in the process of feature extraction.These two aspects are combined and the information of“feature-feature”and“feature-label”is extracted respectively.The comparison experiments on Yahoo high-dimensional multi-label datasets show that the performance of this algorithm is better than the current five main multi-label feature extraction methods,and the effectiveness of the proposed algorithm is verified.