非平衡情感数据背景下基于边界度的过采样方法

An Oversampling Method Based on Boundary Degree in the Background of Unbalanced Affective Data

针对情感分类研究中广泛存在的数据不平衡问题,提出了一种基于边界度的过采样方法(BD-SMOTE)。首先,根据少数类样本的多数类最近邻和少数类最近邻确定其边界度;其次,根据边界度计算少数类样本的采样权重;最后,根据采样权重自适应确定每一个少数类样本需要生成新样本的数量。实验结果表明,将该算法应用于不平衡情感数据集并结合SVM分类器训练分类模型,实现了准确分类。

In view of the widespread data imbalance in the study of emotion classification,a BD-SMOTE oversampling method based on boundary degree is put forward.Firstly,the boundary degree is determined according to the majority nearest neighbor and minority nearest neighbor of minority sample.Secondly,the sampling weights of a few samples are calculated according to the boundary degree.Finally,the number of new samples needed to be generated for each minority sample is determined adaptively according to sampling weight.The experimental results show that the algorithm is applied to the unbalanced emotion data set and combined with SVM classifier to train the classification model,which can achieve accurate classification.