一种改进的RAKEL多标签分类算法

发布时间：2017-12-27 00:07

本文关键词：一种改进的RAKEL多标签分类算法　出处：《浙江师范大学学报(自然科学版)》2016年04期 　论文类型：期刊论文

【摘要】：RAKEL(random k-labelsets)算法是一种集成技术,能有效解决多标签分类问题.它将原始标签集随机选用一小部分标签子集构成的数据集来训练每个分类器,但由于RAKEL算法构造标签空间的随机性,并未充分考察到样本多个标签之间的相关性,从而造成分类精度不高,泛化性能受到一定影响.为此,提出了改进的LC-RAKEL算法.首先,通过标签聚类将原始标签集划分成标签簇,再从每个标签簇中各选择一个标签构成标签集,以此发现标签空间中重要且不频繁的映射关系;然后,利用出现次数较少的标签集合组成新的训练数据,训练相应的分类器.实验证明,改进的算法性能优于其他常用多标签分类算法.
[Abstract]:RAKEL (random k-labelsets) algorithm is a kind of integrated technology, can effectively solve the multi label classification problem. It will be the original label set were randomly selected for a small part of the label subsets of data set to train each classifier, but due to the randomness of the RAKEL algorithm to construct the label space, and not to fully examine the correlation between samples of multiple tags thus, the classification accuracy is not high, the generalization performance will be affected. Therefore, an improved LC-RAKEL algorithm is proposed. Firstly, through the original label tag clustering set into tag clusters, then from each cluster in the choice of a tag tag tag set, in order to find the mapping between the label space is important and not frequent then, the training data set; the new use of fewer labels, training the corresponding classifier. The experimental results show that the improved algorithm has better performance than other commonly used multi label classification Algorithm.
【作者单位】：浙江师范大学数理与信息工程学院;
【分类号】：TP181
【正文快照】： 多标签学习问题在现实世界中广泛存在.例如,在图像分类[1]中,一张图片往往可以对应多个主题,如“海滩”和“落日”.在文档分类[2]中,一篇文档可以属于多个主题,如“世界杯”和“足球”.可以看出,每个样例都与一个标签集合相对应.多标签学习主要研究当样本同时具有多个类别标记

【相似文献】