Abstract: An online reviews mining and requirements elicitation methodology considering product attributes' hierarchy was proposed.Firstly,a large amount of textual reviews was mined by using a semi-supervised method based on hierarchical Latent Dirichlet Allocation(hLDA),and a product attribute hierarchy tree was generated.Then,the emotional polarity words corresponding to the product attribute words were extracted on the basis of the dependency syntax analysis,and were quantitatively expressed in emotional polarity values.Specially,the missing emotional polarity values were supplemented based on the weighted hierarchical relationship among the attributes.Finally,the average user satisfaction value was calculated for each product attribute,and the next-generation product development requirements were analyzed and determined.A case study of mining the word-of-mouth data in the Autohome website to obtain requirements was provided to verify the effectiveness of proposed methodology.

Key words: online reviews, requirements elicitation, hierarchy, product attribute, sentiment polarity

摘要： 为提高需求获取的准确性,提出一种考虑产品属性层次性的评论挖掘及需求获取方法。首先,采用基于层次主题模型的半监督方法挖掘大量文本评论数据,得到产品属性层次结构树。其次,基于依存句法分析抽取产品属性对应的情感极性相关词,并对其量化表示为情感极性值,再基于属性间的加权层次关系对缺失的情感极性值进行补充。最后,计算每个产品属性的平均用户满意度,并分析确定下一代产品的开发需求。通过挖掘汽车之家口碑数据获取用户需求的案例,证明了所提方法的有效性。

关键词: 在线评论, 需求获取, 层次性, 产品属性, 情感极性