计算机科学与探索 ›› 2018, Vol. 12 ›› Issue (3): 341-349.DOI: 10.3778/j.issn.1673-9418.1708061

• 学术研究 • 上一篇    下一篇

中文产品评论的维度挖掘及情感分析技术研究

赵志滨,刘    欢,姚    兰,于    戈+   

  1. 东北大学 计算机科学与工程学院,沈阳 110819
  • 出版日期:2018-03-01 发布日期:2018-03-08

Research on Dimension Mining and Sentiment Analysis for Chinese Product Comments

ZHAO Zhibin, LIU Huan, YAO Lan, YU Ge+   

  1. School of Computer Science and Engineering, Northeastern University, Shenyang 110819, China
  • Online:2018-03-01 Published:2018-03-08

摘要: 产品评论通常会描述产品的多个属性维度,单条评论中所描述的多个维度可能会有不同的维度情感。现实中,用户对产品不同的属性维度的关注度也不相同。反映到评论情感分析中,用户关注度越大的产品维度对评论的整体情感的影响也会越大。细粒度的评论维度挖掘和维度情感分析可以提供很多有价值的市场反馈信息和用户偏好信息。针对电商平台的中文产品评论文本,首先使用规则法抽取产品评论中所描述的维度信息,然后分别针对各个维度计算维度情感。进一步,提出了维度权重计算方法。最后,综合维度情感和维度权重计算评论的整体情感。使用来自于京东商城的真实评论数据集对所提方法进行了综合验证。实验结果表明,所提方法在维度挖掘、维度情感分析、维度权重计算以及整体情感分析方面具有很好的性能。

关键词: 文本挖掘, 维度抽取, 维度权重, 情感分析

Abstract: Generally speaking, product reviews will refer to several dimensions of a product, each of which may have different sentiments. Actually, users have different concerns about the product. The dimensions that users care more should have more influence on the overall sentiment. Fine-grained dimension mining from reviews and dimension-level sentiment analysis can provide a lot of valuable information about market feedback and users?? preference. Unfortunately, many E-commerce platforms are lack of these significant data. This paper aims at mining the dimensions based on rules, along with their sentiments, referred by online product reviews. Next, this paper proposes the method to determine the weights of each product dimension. Then, the overall sentiment of a review can be concluded with the dimension sentiments and their corresponding weights. This paper conducts comprehensive experiments based on the real reviews from JD.com. The experimental results show that the proposed algorithm has a good performance in dimension extraction, dimension sentiment mining, weight calculation and overall sentiment determination.

Key words:  text mining, dimension mining, dimension weight, sentiment analysis