第15卷第2期 智能系统学报 Vol.15 No.2 2020年3月 CAAI Transactions on Intelligent Systems Mar.2020 D0:10.11992/tis.201810014 网络出版地址：http:/kns.cnki.net/kcms/detail/23.1538.TP.20190829.0906.002.html 面向众包数据的特征扩维标签质量提高方法 李易南，王士同 （江南大学数字媒体学院，江苏无锡214122) 摘要：众包是一个新兴的收集数据集标签的方法。虽然它经济实惠，但面临着数据标签质量无法保证的问 题。尤其是当客观原因存在使得众包工作者工作质量较差时，所得的标签会更加不可靠。因此提出一个名为 基于特征扩维提高众包质量的方法(FA-method),其基本思想是，首先由专家标注少部分标签，再利用众包者标 注的数据集训练模型，对专家集进行预测，所得结果作为专家数据集新的特征，并利用扩维后的专家集训练模 型进行预测，计算每个实例为噪声的可能性以及噪声数量上限来过滤出潜在含噪声标签的数据集，类似地，对 过滤后的高质量集再次使用扩维的方法进一步校正噪声。在8个UCI数据集上进行验证的结果表明，和现有 的结合噪声识别和校正的众包标签方法相比，所提方法能够在重复标签数量较少或标注质量较低时均取得很 好的效果。 关键词：众包：标签质量：扩维：专家标注：噪声识别：噪声校正：噪声可能性：噪声数量上限 中图分类号：TP181文献标志码：A文章编号：1673-4785(2020)02-0227-08 中文引用格式：李易南，王士同.面向众包数据的特征扩维标签质量提高方法.智能系统学报，2020,15(2)：227-234. 英文引用格式：LI Yinan,WANG Shitong..A feature augmentation method for enhancing the labeling quality of crowdsourcing data[J].CAAI transactions on intelligent systems,2020,15(2):227-234. A feature augmentation method for enhancing the labeling quality of crowdsourcing data LI Yinan,WANG Shitong (School of Digital Media.Jiangnan University,Wuxi 214122,China) Abstract:Crowdsourcing is a new method of collecting the labels of data.Although it is economical,crowdsourcing faces an unavoidable problem,i.e.,the quality of the labels cannot be guaranteed.In particular,when the quality of la- beling work is low because of the existence of objective causes,the result of crowdsourcing will be unreliable.In this study,a feature augmentation method for enhancing the labeling quality of crowdsourcing data is proposed.In the pro- posed method,first,a small amount of expert data is labeled by several people with professional knowledge.Then,the crowdsourcing data are used to create the classifiers and predict the expert data.The resultant predicted labels are used to augment the expert data.Then,the augmented expert data are used to create the classifiers,predict the original data, and calculate the probability of noise for each instance and the upper limit of noise number to filter out the high-quality dataset from potentially noisy labels.Similarly,the filtered high-quality dataset is utilized to further correct the noisy la- bels using the proposed feature augmentation method.The experiments conducted on eight UCI datasets show that the proposed feature augmentation method has achieved encouraging results when the number of repeated labels is compar- atively small or the quality of labeling is comparatively low. Keywords:crowdsourcing;labeling quality;feature augmentation;expert labeling;noise identification;noise correc- tion;noise probability;upper limit of noise number 收稿日期：2018-10-15.网络出版日期：2019-08-29. 基金项目：国家自然科学基金项目(61272210). 众包，一般认为是由美国作家杰夫·豪(Jeff 通信作者：李易南.E-mail:1920898036@qq.com Howe)在2006年6月于《众包：大众力量缘何推DOI: 10.11992/tis.201810014 网络出版地址: http://kns.cnki.net/kcms/detail/23.1538.TP.20190829.0906.002.html 面向众包数据的特征扩维标签质量提高方法 李易南，王士同 （江南大学 数字媒体学院，江苏 无锡 214122） 摘 要：众包是一个新兴的收集数据集标签的方法。虽然它经济实惠，但面临着数据标签质量无法保证的问 题。尤其是当客观原因存在使得众包工作者工作质量较差时，所得的标签会更加不可靠。因此提出一个名为 基于特征扩维提高众包质量的方法 (FA-method)，其基本思想是，首先由专家标注少部分标签，再利用众包者标 注的数据集训练模型，对专家集进行预测，所得结果作为专家数据集新的特征，并利用扩维后的专家集训练模 型进行预测，计算每个实例为噪声的可能性以及噪声数量上限来过滤出潜在含噪声标签的数据集，类似地，对 过滤后的高质量集再次使用扩维的方法进一步校正噪声。在 8 个 UCI 数据集上进行验证的结果表明，和现有 的结合噪声识别和校正的众包标签方法相比，所提方法能够在重复标签数量较少或标注质量较低时均取得很 好的效果。 关键词：众包；标签质量；扩维；专家标注；噪声识别；噪声校正；噪声可能性；噪声数量上限 中图分类号：TP181 文献标志码：A 文章编号：1673−4785(2020)02−0227−08 中文引用格式：李易南, 王士同. 面向众包数据的特征扩维标签质量提高方法 [J]. 智能系统学报, 2020, 15(2): 227–234. 英文引用格式：LI Yinan, WANG Shitong. A feature augmentation method for enhancing the labeling quality of crowdsourcing data[J]. CAAI transactions on intelligent systems, 2020, 15(2): 227–234. A feature augmentation method for enhancing the labeling quality of crowdsourcing data LI Yinan，WANG Shitong (School of Digital Media, Jiangnan University, Wuxi 214122, China) Abstract: Crowdsourcing is a new method of collecting the labels of data. Although it is economical, crowdsourcing faces an unavoidable problem, i.e., the quality of the labels cannot be guaranteed. In particular, when the quality of la￾beling work is low because of the existence of objective causes, the result of crowdsourcing will be unreliable. In this study, a feature augmentation method for enhancing the labeling quality of crowdsourcing data is proposed. In the pro￾posed method, first, a small amount of expert data is labeled by several people with professional knowledge. Then, the crowdsourcing data are used to create the classifiers and predict the expert data. The resultant predicted labels are used to augment the expert data. Then, the augmented expert data are used to create the classifiers, predict the original data, and calculate the probability of noise for each instance and the upper limit of noise number to filter out the high-quality dataset from potentially noisy labels. Similarly, the filtered high-quality dataset is utilized to further correct the noisy la￾bels using the proposed feature augmentation method. The experiments conducted on eight UCI datasets show that the proposed feature augmentation method has achieved encouraging results when the number of repeated labels is compar￾atively small or the quality of labeling is comparatively low. Keywords: crowdsourcing; labeling quality; feature augmentation; expert labeling; noise identification; noise correc￾tion; noise probability; upper limit of noise number 众包，一般认为是由美国作家杰夫·豪 (Jeff Howe) 在 2006 年 6 月于《众包：大众力量缘何推 收稿日期：2018−10−15. 网络出版日期：2019−08−29. 基金项目：国家自然科学基金项目 (61272210). 通信作者：李易南. E-mail：1920898036@qq.com. 第 15 卷第 2 期 智 能 系 统 学 报 Vol.15 No.2 2020 年 3 月 CAAI Transactions on Intelligent Systems Mar. 2020