基于标签相关性的多标签分类算法  

Multi-label Classification Algorithm Based on Label Correlation

作　　者：梁义珂 谢小兰[1] 张迎春[2] 张珣[1] LIANG Yi-ke;XIE Xiao-lan;ZHANG Ying-chun;ZHANG Xun(College of Computer,Beijing Technology and Business University,Beijing 100048,China;Computer Network Center,Beijing Technology and Business University,Beijing 100048,China)

机构地区：[1]北京工商大学计算机学院,北京100048 [2]北京工商大学计算机网络中心,北京100048

出　　处：《计算机仿真》2021年第12期255-261,共7页Computer Simulation

基　　金：“十三五”时期北京市属高校高水平教师队伍建设支持计划(CIT&TCD201904037);中国博士后科学基金第62批面上资助(2017M620885)。

摘　　要：多标签分类算法已广泛应用于文本分类、图像识别、基因功能分类等领域,为了解决多标签分类算法较少考量标签之间的相关性等问题,提出一种基于标签相关性的多标签分类算法。首先对BP神经网络进行改进以适应多标签分类算法,然后对标签集分别使用皮尔逊相关系数和关联规则进行二阶、高阶相关性分析,最后将标签的相关性与改进的BP神经网络算法得到的概率做线性插值,得到样本属于某标签的最终概率。通过在4个真实数据集上利用5个多标签分类指标做对比实验,验证了提出的算法分类效果明显优于现有的多标签分类算法。Multi-label classification algorithm has been widely used in text classification, image recognition,gene functional classification and other fields. In order to solve the problem that multi-label classification algorithm seldom considers the correlation between labels, this paper proposes a multi-label classification algorithm based on label correlation. Firstly, the BP neural network was improved to adapt to the multi-label classification algorithm.Then we respectively used Pearson correlation coefficient and association rules to analyze the second-order and highorder correlation of the label set. Finally, we linearly interpolated the correlation between the two label and the multiple label with the improved BP neural network algorithm to obtain the final probability that this sample belonged to some label. Based on the comparison of five multi-label classification indicators on four real data sets, it is verified that the proposed algorithm is better than the existing multi-label classification algorithm.

关 键 词：多标签分类 神经网络 皮尔逊相关系数 关联规则 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]