摘要
App分类是App检索、App推荐和用户偏好挖掘等智能服务的前提。现有App下载平台已提供了App的类别信息,但这些信息存在无法被第三方使用、无法满足定制的类别体系等问题,因此需要设计一种不依赖App下载平台的App分类方法,而要实现这一目标的挑战包括能利用的描述信息太少(通常仅有App的名称)和有标注样本的数量太少(特别是需要定制类别体系的时候)。针对这些挑战,提出了一种基于多视图协同学习的App分类方法:首先,基于互联网知识对App名称进行信息扩充;然后,从词频和主题两个视图描述App分类问题,并抽取词频分布特征和主题分布特征;最后,采用协同学习方法充分利用无标注样本来训练和融合基于这两个视图的分类器。基于大量真实App数据的实验表明:相比现有方法,提出的方法在分类准确度方面有一定优势。
App classification is the requirement of many intelligent services,e.g.,App query,App recommendation,user preference mining,etc.Although many App delivery platforms have already provided the category information of the Apps,this information can be not used by third parties and cannot adapt to customized taxonomy.Thus,we need an App classification method,which is independent of the App delivery platforms.The main challenges include limited descriptive information(only the name of an App is available)and limited labeled samples(especially when customized taxonomy is required).Aiming at these challenges,this paper proposes an App classification method based on multi-view collaborative learning.Firstly,it enriches the descriptive information through exploiting the additional Web knowledge.Secondly,it represents the App classification problem from two views(i.e.,word frequency view and latent topic view),and extracts word frequency distribution features and latent topic distribution features respectively.Finally,it fully utilizes unlabeled samples to train and fuse the classifiers based on these two views using a collaborative learning approach.The experiments based on real App data show that the proposed method has a competitive App classification accuracy over state-of-the-art methods.
作者
陈铁明
张凯一
吕明琪
CHEN Tieming;ZHANG Kaiyi;Lü Mingqi(College of Computer Science and Technology,Zhejiang University of Technology,Hangzhou 310023,China)
出处
《浙江工业大学学报》
CAS
北大核心
2018年第6期591-597,共7页
Journal of Zhejiang University of Technology
基金
国家自然科学基金资助项目(61772026)
国家自然科学基金浙江两化融合联系基金项目(U1509214)
浙江省自然科学基金资助项目(LY18F020033)
关键词
App分类
互联网知识
多视图学习
协同学习
App classification
web knowledge
multi-view learning
collaborative learning