基于LDA主题模型的用户特征预测研究

Research on User Traits Predicting Based on LDA Topic Model

用户特征可以通过在线用户的点赞信息进行奇异值分解和Logistic回归有效预测,然而对新用户的特征预测却难以实现。为了解决该问题,提出了一种基于LDA主题模型的在线用户特征预测方法。首先使用LDA模型提取微博用户的点赞文本主题,然后基于主题对新用户的特征进行预测,最后与基于奇异值分解的传统方法比较预测结果。实验结果表明其F1值最高提升0.15,且计算时间平均缩短了69.09%。研究改进了点赞信息固有标签不能准确反映用户偏好的缺陷,避免了传统方法预测过程中仍需对新用户及其点赞信息重新计算的繁琐弊端,为用户特征分析提供了另一条可行途径。

User traits can be effectively predicted by singular value decomposition and Logistic Regression through online user’s‘Like’information.However,this method cannot predict new users’traits.To slove the problem,this paper proposes an online user traits predicting method based on LDA topic model.Firstly,the method extracted the Weibo user’s‘Like’text topic through LDA model.Then it predicted new user traits based on topic.Finally,the result is compared to the traditional method based on singular value decomposition.The results showed that the F1 value of this method was up to 0.15,and the calculation time was shortened by 69.09%in average.Research inproves the defect that the inherent tags of the‘Like’informations cannot accurately reflect user preference,avoiding the disadvantage of recalculating new users and their‘like’information in the predicting process of traditional methods,providing another feasible way for user traits analysis.