基于加权SVM主动学习的多标签分类.pdf

第37 卷 第8 期 计 算 机 工 程 2011 年4 月 Vol.37 No.8 Computer Engineering April 2011 ·人工智能及识别技术· 文章编号：1000—3428(2011)08—0181—02 文献标识码：A 中图分类号：TP18 基于加权SVM 主动学习的多标签分类 刘端阳，邱卫杰 (浙江工业大学计算机科学与技术学院，杭州 310023) 摘 要：样本标记是一个重要但又比较耗时的过程。得到一个多标签分类器需要大量的训练样本，而手工为每个样本创建多个标签会存在 一定困难。为尽可能降低标记样本的工作量，提出一种加权决策函数的主动学习方法，该方法同时考虑训练样本的数量和未知样本的置信 度，使得分类器能在最小的成本下最快地达到比较满意的分类精度。 关键词：主动学习；多标签；支持向量机；训练样本 Multi-label Classification Based on Weighted SVM Active Learning LIU Duan-yang, QIU Wei-jie (College of Computer Science and Technology, Zhejiang University of Technology, Hangzhou 310023, China) 【Abstract 】Manually creating multiple labels for each sample is very important but it is time-consuming. Manually creating multiple labels for each sample may become impractical when a very large amount of data is needed for training multi-label classifier. To minimize the human-labeling efforts, this paper proposes a weighted decision approach, the approach considers quantity and confidence of training samples, it can make the classifier need fewer samples, but achieve a comparative precision. 【Key words 】active learning; multi-label; Support Vector Machine(SVM); training samples DOI ：10.3969/j.issn.1000-3428.2011.08.062 标签集的表示方式为 Y =[y 1, y 2 …,y n] 。用y j ←{-1, 1}来表示 1 概述 i i i i i