Mining user preferences from online reviews to understand the representative preferences of different customer groups plays a critical role in product development and improvement, especially in personalized product design. Previous research on mining user preferences usually assumes that all consumers' preferences are homogenous and does not take differences in consumers’ personalities into account. Besides, traditional online review deep mining methods are too broad to focus on precise and detailed mining of customer preferences. To fill the gaps in existing research, our study develops a template matching deep mining method to segment customers and narrow the mining scope of customer group preference, and then proposes a product family lean improvement model. Firstly, K-means and structural change model are applied to cluster customers reliably based on the similarity of user preferences. Secondly, in order to decrease down mining scope of customer group preference, the Improved Deep Structured Semantic Model is designed to determine sentimental polarity sentimental polarity of different groups by matching the standard sentimental polarity review templates and online reviews. Finally, a KANO mapping model is developed to decide the user preferences for product attributes in each customer group according to their sentimental polarity and further summarize the common preferences and personalized preferences of various groups according to the Preference Commonality Measurement Function. Accordingly, product family lean improvement strategies are proposed to provide product developers with improvement directions. An empirical study is carried out on laptop data on JD.COM to verify the validity of the proposed model and product family lean improvement suggestions are put forward.

从在线评论中挖掘用户偏好以了解不同客户群体的代表偏好在产品开发和改进（尤其是个性化产品设计）中起着至关重要的作用。先前关于挖掘用户偏好的研究通常假定所有消费者的偏好是同质的，并且没有考虑消费者个性的差异。此外，传统的在线评论深度挖掘方法过于广泛，以致于无法专注于准确，详细地挖掘客户的偏好。为了填补现有研究的空白，我们的研究开发了一种模板匹配的深度挖掘方法，以细分客户并缩小客户群偏好的挖掘范围，然后提出了产品系列精益改进模型。首先，基于用户偏好的相似性，将K均值和结构更改模型可靠地应用于集群客户。其次，为了缩小客户群偏好的挖掘范围，设计了改进的深度结构化语义模型，通过匹配标准的情感极性评论模板和在线评论来确定不同组的情感极性。最后，开发了KANO映射模型，以根据每个客户组的情感极性确定用户对产品属性的偏好，并根据偏好共性度量功能进一步总结各个组的共同偏好和个性化偏好。因此，提出了产品系列精益改进策略，以为产品开发人员提供改进方向。