系统管理学报

系统管理学报 ›› 2021, Vol. 30 ›› Issue (3): 481-489.DOI: 10.3969/j.issn.1005-2542.2021.03.007

• 管理信息系统 • 上一篇    下一篇

基于混合类别序列规则的中文比较评论的识别

朱茂然，蒋凯艳，高松，王洪伟   

  1. 1.同济大学 经济与管理学院，上海 200092；2.中国信息安全测评中心，北京100085
  • 出版日期:2021-05-28 发布日期:2021-06-16
  • 通讯作者: 王洪伟（1973-），男，博士，教授，博士生导师
  • 作者简介:朱茂然（1972-），男，博士，副教授。研究方向为文本挖掘和IT审计
  • 基金资助:
    国家自然科学基金资助项目（71771177，71701085）；教育部科技发展中心高校产学研创新基金并且项目（2019J01012）；中国标准化协会服务贸易标准化科研课题（FMBZH-1947）

Identification of Comparative Information from Chinese Online Reviews Based on a Hybrid Class Sequential Rules Method

ZHU Maoran，JIANG Kaiyan，GAO Song，WANG Hongwei   

  1. 1. School of Economics and Management，Tongji University，Shanghai 200092，China；2. China Information Technology Security Evaluation Center，Beijing 100085，China
  • Online:2021-05-28 Published:2021-06-16

PDF

摘要： 在线评论中的比较信息揭示了品牌和商品的竞争性关系，为消费者的购买决策提供了有力的依据。在线评论中比较信息通常以显性和隐性两种形式存在。为此，提出一种将句法、规则、特征相结合的比较句识别算法。针对显性比较评论，提出了融合CSR方法与依存句法分析算法，即比较句的形式化结构与内在依存关系两方面结合，提高比较句识别的效率。针对隐性比较评论，提出了基于产品名识别的方法，可有效识别隐性比较句，由此拓宽了比较句识别的范围。以京东购物平台为数据来源，实验表明，本文方法在比较关系识别上获得了较好的效果。

关键词: 比较句识别, 类别序列规则, 依存句法, 隐性比较关系

Abstract: Comparative information found in online reviews can reveal the competitive relations between brands and commodities, providing a strong basis for consumers to make purchase decisions. Usually, comparative information exists either in an explicit way or in an implicit way. Thus, this paper intends to propose a hybrid algorithm for identifying comparative sentences, which combines syntax, rules, and features. In order to identifying explicit comparison comments, it raises an algorithm combining the CSR algorithm and the dependency parsing method. The proposed algorithm enables the improvement of efficiency for comparative sentence recognition by comparing structure and syntax between sentences. Regarding implicit comparison comments, it proposes a method based on product name recognition to effectively identify the implicit comparison sentences by using the data from the online reviews in JD.com website. The experiment results show that the proposed method can achieve significant improvement in identifying comparative online reviews.