用户级情感预测主题模型的构建与研究

摘要

The emotion triggered by document topics is associated with user features.To improve the accuracy of emo-tion prediction by exploiting user features,a layer is added to sample user information for existing two-level topic models MSTM and SLTM.For the respectively proposed three-level topic models UMSTM and USLTM,user features,document topics,and emotions are modeled jointly.To validate the effectiveness of UMSTM and USLTM,experiments over the pro-posed models and baselines are conducted using metrics of accuracy and correlation coefficient.The experimental results show that UMSTM outperforms MSTM in terms of accuracy and correlation coefficient.Furthermore,USLTM achieves better performance than SLTM in both evaluation metrics.%文本主题引发的情感反馈与用户特征之间具有一定的关联.为了充分挖掘用户特征的价值以提高情感预测的准确度,在双层主题模型MSTM和SLTM的基础上,增加了对用户特征信息的采样层,进而提出了基于用户特征的"用户-主题-情感"三层主题模型UMSTM和USLTM.通过三层模型与基础模型在最高情感命中率以及情感概率预测相关系数的对比实验,来检验用户特征对情感预测产生的效果与影响.实验验证了UMSTM和USLTM在以上两种指标中,相对于MSTM和SLTM均有提高.