计算机工程与应用 ›› 2017, Vol. 53 ›› Issue (15): 149-154.DOI: 10.3778/j.issn.1002-8331.1611-0129

• 模式识别与人工智能 • 上一篇    下一篇

融合多样性测度的汉语方言主动辨识方法

夏玉果1,戴红霞1,顾明亮2   

  1. 1.江苏信息职业技术学院 电子信息工程学院,江苏 无锡 214153
    2.江苏师范大学 语言科学学院,江苏 徐州 221116
  • 出版日期:2017-08-01 发布日期:2017-08-14

Chinese dialect active identification method fusing diversity measure

XIA Yuguo1, DAI Hongxia1, GU Mingliang2   

  1. 1.School of Electronic Information Engineering, Jiangsu Vocational College of Information Technology, Wuxi, Jiangsu 214153, China
    2.School of Linguistic Science, Jiangsu Normal University, Xuzhou, Jiangsu 221116, China
  • Online:2017-08-01 Published:2017-08-14

摘要: 为了解决方言辨识系统中训练样本冗余的问题,提出了一种融合多样性测度的汉语方言主动辨识方法。利用SVM分类器选取不确定性的样本。根据样本间分布情况的测度算法,选取出兼具多样性的训练样本,经过多次迭代将这些最具区别性的样本组成训练集。将此训练集重新输入到SVM进行分类辨识。实验结果表明,该方法能有效克服选取样本的冗余,与传统的主动学习方法相比,在同等识别率的情况下,人工标注样本的数量减少了50%。

关键词: 汉语方言辨识, 主动学习, 支持矢量机, 多样性测度

Abstract: In order to solve the problem of the redundant training samples in dialect identification system, an approach for Chinese dialect identification fusing diversity measure is proposed. Firstly, the uncertain samples are chosen by SVM classifier, then according to the distribution of these samples, the uncertain samples with diversity are selected and the new training set including these distinctive samples is constructed after several iterations. Finally, SVM is reused to make the decision. Experimental results indicate that, compared with the traditional active method, the proposed approach effectively overcomes the redundancy of the samples and the number of manually annotated samples reduces 50% under the same condition of recognition accuracy.

Key words: Chinese dialect identification, active learning, Support Vector Machine(SVM), diversity measure