【作者】 汪杨；

【导师】 王刚； 周保昌；

【作者基本信息】 合肥工业大学 ， 物流工程（专业学位）， 2016， 硕士

【摘要】 信息技术的飞速发展和互联网技术的全面普及增强了用户收集、存储和传输数据的能力,与此同时也推动了“大数据时代”的到来。大数据当中蕴含丰富的信息,但同时也呈现出体量大和类型多等特征,这也就造成了难于处理和信息过载等问题。为了从海量的数据中获取到满足需求的指定信息,数据挖掘技术应运而生。而多标记学习又是当前数据挖掘领域的一个研究热点,其可以有效的解决实际应用中的多标记分类等问题,因此,多标记学习已受到学术界和产业界的高度重视。目前,已经有许多研究者对多标记学习进行了系统的研究,提出了很多的多标记学习方法。但这些已有方法在应用中仍然存在一定问题,其中一个重要问题就是高维数据问题。为此,本研究将基于特征选取的Random Subspace方法引入到多标记学习当中,构建基于Random Subspace的分类器链方法RS-CC和基于Random Subspace的组合分类器链方法RS-ECC。首先,本文系统分析了多标记学习的研究现状,明确了当前的研究问题和未来的研究方向。其次,对多标记学习的相关基础理论进行了系统研究,理解了机器学习、监督学习和多标记学习等知识模块。然后,针对高维数据条件下的多标记学习问题,考虑到分类器链方法的不稳定性和组合分类器链方法在解决高维数据问题时的算法复杂程度高等劣势,从特征提取的角度分别构建基于Random Subspace的分类器链方法RS-CC和基于Random Subspace的组合分类器链方法RS-ECC,并通过多个标准数据集对其有效性进行了检验。最后,将面向高维数据的改进多标记链式学习方法RS-CC和RS-ECC应用于物流专家推荐这一高维并且多义的分类问题中,通过从科研之友社交网站上抓取的物流专家数据集对改进方法的有效性进行了检验。实验结果表明,面向高维数据的改进多标记链式学习方法RS-CC和RS-ECC在多个应用领域都取得了较好的分类结果,能够很好的处理高维数据多标记分类问题,从而验证了本研究提出的改进方法的有效性。通过本研究,一方面对多标记学习领域的相关理论进行了系统的分析,提出了改进的多标记学习方法,丰富和完善了多标记学习的理论研究体系；另一个方面,本研究率先将多标记学习方法应用到物流领域的专家推荐当中,拓展了多标记学习的应用范围,与此同时,也为物流专家推荐提供了一个新的解决问题的思路和途径。更多还原

【Abstract】 The rapid development of information technology has brought about the "big data era". Big data contains a large scale of information, which results in information overload. Data mining technologies have been introduced to help get specific information from the mass data. Multi-label learning is a research hotspot in data mining, which can effectively solve the practical application of classification. Therefore, multi-label learning received more and more attention nowadays.Currently, researchers have focused on theory and application of multi-label learning. They have proposed several multi-label learning algorithms. However, challenges in applying them in different applications domains still exist. One major challenge is the high-dimensional data. In order to solve this challenge, this research proposed new multi-label learning algorithms based on Random subspace. First, research status on multi-label learning was analyzed, which clarifies the research problems and future directions. Secondly, the basic theory of multi-label learning, including machine learning, supervised learning, and multi-label learning were analyzed. Finally, for instability of classification chain and the high complexity of ensemble classification chains, this research constructed a new RS-CC and RS-ECC methods. The proposed methods were applied into logistics experts recommendation. Experimental results indicated that proposed methods achieved good results.In summary, this research made contributions to both theory and practice. On one hand, the theory of multi-label learning has been analyzed and enriched. On the other hand, this research applied the multi-label learning algorithms based on random subspace in logistics experts recommendation, which has expanded the application range of multi-label learning, and provided a new method for logistics experts recommendation.更多还原