面向电子商务网站的产品属性提取算法  

Algorithm for Extracting Product Attributes from E-commerce Websites

作　　者：李俊[1] 陈黎[1] 王亚强[1] 秦湘清[1] 于中华[1] 

机构地区：[1]四川大学计算机学院,成都610065

出　　处：《小型微型计算机系统》2013年第11期2477-2481,共5页Journal of Chinese Computer Systems

基　　金：高等学校博士学科点专项科研基金项目(20100181120029)资助

摘　　要：从商品评论中抽取作为评价对象的产品属性及判断评价的极性(正面评价、负面评价、中性评价),对于充分挖掘利用电子商务网站上积累的大量商品评论,为消费者的购物决策和生产者的生产决策提供支持,具有重要意义.本文针对现有算法的不足,结合中文电子商务网站中商品评论的特点,提出了综合模板、频率和HITS的无监督学习算法,用于从中文商品评论中识别产品属性.充分的实验结果表明,所提出的无监督算法对产品属性识别的F值可以达到77.3%,优于文献中提出的其他类似算法.Extracting product attributes which are treated as comment targets from product review and determining the opinion orienta-tion （ positive, negative and neutral ） are of great significance for fully mining a large number of product reviews on e-commerce web-site and providing support for shopping decision, production decision of consumers and producers respectively. In view of the defi-ciency of existing approaches and combining with the characteristics of the product reviews in Chinese e-commerce sites, this paper proposes an unsupervised method which integrates POS patterns, term frequency and HITS to identify product attributes from Chinese product reviews. The experimental results show that the proposed method can reach an F-score of 77.3 % on product attributes extrac- tion, outperform the existing other methods.

关 键 词：商品评论 产品属性 抽取 HITS 抽取模板 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]