数据流分类是数据流挖掘领域一项重要研究任务,目标是从不断变化的海量数据中捕获变化的类结构.目前,几乎没有框架可以同时处理数据流中常见的多类非平衡、概念漂移、异常点和标记样本成本高昂问题.基于此,提出一种非平衡数据流在线主...数据流分类是数据流挖掘领域一项重要研究任务,目标是从不断变化的海量数据中捕获变化的类结构.目前,几乎没有框架可以同时处理数据流中常见的多类非平衡、概念漂移、异常点和标记样本成本高昂问题.基于此,提出一种非平衡数据流在线主动学习方法(Online active learning method for imbalanced data stream,OALM-IDS).AdaBoost是一种将多个弱分类器经过迭代生成强分类器的集成分类方法,AdaBoost.M2引入了弱分类器的置信度,此类方法常用于静态数据.定义了基于非平衡比率和自适应遗忘因子的训练样本重要性度量,从而使AdaBoost.M2方法适用于非平衡数据流,提升了非平衡数据流集成分类器的性能.提出了边际阈值矩阵的自适应调整方法,优化了标签请求策略.将概念漂移程度融入模型构建过程中,定义了基于概念漂移指数的自适应遗忘因子,实现了漂移后的模型重构.在6个人工数据流和4个真实数据流上的对比实验表明,提出的非平衡数据流在线主动学习方法的分类性能优于其他5种非平衡数据流学习方法.展开更多
Hundreds of scientific journals are published in China. However, only scores of them are included in Science Citation Index by the Institute for Scientific Information, with impact factors of only 1 or less. Thus, how...Hundreds of scientific journals are published in China. However, only scores of them are included in Science Citation Index by the Institute for Scientific Information, with impact factors of only 1 or less. Thus, how to establish a first-class international scientific journal in China is an important but difficult topic that deserves extensive exploration. World Journal of Gastroenterology (WJG) sets a good example although it has experienced setbacks on the road towards success. Concepts and pursuits that affirm the overall development direction, innovation and dreams that provide impetus and aspiration for higher objectives, team work and unique pattern that assure excellent quality and service, and culture and environment that also determine the speed and direction of the development, are believed to be the major factors contributing to the success of WJG. It is recommended that the effective resolution to the above issue is to learn from Chinese examples such as WJG rather than from “how foreign journals do”.展开更多
A new classification algorithm for web mining is proposed on the basis of general classification algorithm for data mining in order to implement personalized information services. The building tree method of detecting...A new classification algorithm for web mining is proposed on the basis of general classification algorithm for data mining in order to implement personalized information services. The building tree method of detecting class threshold is used for construction of decision tree according to the concept of user expectation so as to find classification rules in different layers. Compared with the traditional C4.5 algorithm, the disadvantage of excessive adaptation in C4.5 has been improved so that classification results not only have much higher accuracy but also statistic meaning.展开更多
文摘数据流分类是数据流挖掘领域一项重要研究任务,目标是从不断变化的海量数据中捕获变化的类结构.目前,几乎没有框架可以同时处理数据流中常见的多类非平衡、概念漂移、异常点和标记样本成本高昂问题.基于此,提出一种非平衡数据流在线主动学习方法(Online active learning method for imbalanced data stream,OALM-IDS).AdaBoost是一种将多个弱分类器经过迭代生成强分类器的集成分类方法,AdaBoost.M2引入了弱分类器的置信度,此类方法常用于静态数据.定义了基于非平衡比率和自适应遗忘因子的训练样本重要性度量,从而使AdaBoost.M2方法适用于非平衡数据流,提升了非平衡数据流集成分类器的性能.提出了边际阈值矩阵的自适应调整方法,优化了标签请求策略.将概念漂移程度融入模型构建过程中,定义了基于概念漂移指数的自适应遗忘因子,实现了漂移后的模型重构.在6个人工数据流和4个真实数据流上的对比实验表明,提出的非平衡数据流在线主动学习方法的分类性能优于其他5种非平衡数据流学习方法.
文摘Hundreds of scientific journals are published in China. However, only scores of them are included in Science Citation Index by the Institute for Scientific Information, with impact factors of only 1 or less. Thus, how to establish a first-class international scientific journal in China is an important but difficult topic that deserves extensive exploration. World Journal of Gastroenterology (WJG) sets a good example although it has experienced setbacks on the road towards success. Concepts and pursuits that affirm the overall development direction, innovation and dreams that provide impetus and aspiration for higher objectives, team work and unique pattern that assure excellent quality and service, and culture and environment that also determine the speed and direction of the development, are believed to be the major factors contributing to the success of WJG. It is recommended that the effective resolution to the above issue is to learn from Chinese examples such as WJG rather than from “how foreign journals do”.
文摘A new classification algorithm for web mining is proposed on the basis of general classification algorithm for data mining in order to implement personalized information services. The building tree method of detecting class threshold is used for construction of decision tree according to the concept of user expectation so as to find classification rules in different layers. Compared with the traditional C4.5 algorithm, the disadvantage of excessive adaptation in C4.5 has been improved so that classification results not only have much higher accuracy but also statistic meaning.
