基于多路层次聚类的商品评论数据概念分类构建  

Multi-Way Hierarchical Clustering Based Concept Taxonomy Construction for Product Reviews

作　　者：涂鼎 陈岭[1] 陈根才[1] 吴勇[2] 王敬昌[2] 

机构地区：[1]浙江大学计算机科学与技术学院,杭州310027 [2]浙江鸿程计算机系统有限公司,杭州310009

出　　处：《计算机研究与发展》2013年第S2期208-215,共8页Journal of Computer Research and Development

基　　金：国家“核高基”重大科技专项课题(2010ZX01042-002-003);国家自然科学基金项目(60703040);浙江省重大科技专项(2011C13042,2013C01046)

摘　　要：针对基于传统层次聚类构建概念分类只能生成二叉树形式结果的限制,提出一种新的评论集概念分类构建方法.首先,使用主题模型对评论集进行描述,选出评论集中最具代表性的主题词作为候选概念词.接着,使用WordNet提取概念词间的语义关系,并计算概念词间的语义距离.最后,对概念词进行多路层次聚类,生成概念分类.此外,还提出一种新的概念分类构建性能评价标准,并使用该标准在用户评论数据集上将提出方法与基本层次聚类方法进行了比较,结果表明使用多路层次聚类的方法能够获得与目标概念分类更高的相似度.Traditional hierarchical clustering based methods construct a concept taxonomy in a binarytree form.To relieve this constraint,a new concept taxonomy construction method for product reviews is proposed in this paper.First,a topic model is applied to describe the review corpus and the most representative words are extracted.Then,the semantic distances between these words are computed based on WordNet.Finally,all concept words are clustered into a concept taxonomy using a multi-way hierarchical clustering algorithm.Besides that,a new evaluation metric is presented,and the proposed method is compared with basic hierarchical clustering methods on a customer review data set under this metric.The result shows that the method using multi-way hierarchical clustering gets higher similarity with the referenced concept taxonomy.

关 键 词：概念分类 WORDNET 层次聚类 文本挖掘 本体评价 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]