Abstract: Considering the horizontal correlation and vertical hierarchy of product attributes in the short text data of cloud service platform,a dynamic method for acquiring product attributes of hLAT users' preference was proposed.The hierarchical Latent Dirichlet allocation (hLDA) model was used to mine the on-line text topic hierarchy and construct the initial product attribute hierarchy tree.The influence of word topic was considered to modify the random jump probability between nodes to optimize the TextRank algorithm,An example of automobile industry related problem data in the platform of "Orange Cloud Industrial Product Collaborative Research and development" was given,and the feasibility and effectiveness of the proposed method were verified,which provided a new idea for the front-end construction of cloud service platform.

Key words: cloud services, online text, user preferences, product attributes, hLDA, TextRank

摘要： 考虑云服务平台短文本数据中横向关联性与产品属性纵向层次性,提出一种层次主题模型（hLAT）用户偏好产品属性动态获取方法。首先,利用hLDA模型纵向挖掘在线文本主题层次结构,构建初始产品属性层次树;其次,考虑词语主题影响力因素以修改节点间的随机跳转概率优化TextRank算法,获取候选偏好词排序构建用户偏好产品属性主题层次树;最后,以“橙色·云工业产品协同研发”平台中“汽车工业”相关问题数据为例,验证了所提方法的可行性和有效性,为云服务平台前端建设提供了崭新构思。

关键词: 云服务, 在线文本, 用户偏好, 产品属性, 层次主题模型, 词图模型