Abstract: Product reviews mining is mainly used to extract the information from amount of consumers’ comments on the Web，and then the consumers’ positive or negative semantic orientation on a product’s parts or function is gotten.The existing product reviews mining do not deal with either hyponymy features or the same feature described in different words.This article firstly mines the manufacturers’ specification files to get the specification features and their hierarchies，and then uses the Bootstrapping weakly supervisory method to extract the consumers’ describing features from the website editors’ evaluating articles and gets its hierarchies with specification features.This method has been used in mobile phone features research.The experiment shows that the hierarchies among a product’s features extracted in this way are efficient.

Key words: data mining, reviews mining, Bootstrapping, textual pattern extraction

摘要： 产品评论挖掘用来对用户发表到网络上的众多评论内容进行信息提取，从而获得用户对产品的部件或功能的褒贬评价。现有的产品评论挖掘研究中没有对上下位的特征、同一特征的不同词语表达进行处理。首先对厂家规格说明文档的结构化表示进行挖掘获得厂家规格特征及其关系，再使用Bootstrapping弱监督方法从网站编辑评测文章中抽取出用户的描述特征及与规格特征之间的层次关系。应用该方法在手机领域的产品特征关系进行了抽取，实验结果显示获得的产品特征之间的层次关系很好的效果。

关键词: 数据挖掘, 评论挖掘, Bootstrapping, 文本模式抽取