结合语义相似度改进LDA的文本主题分析  

Modified LDA model based on semantic similarity for topic analysis of text

作　　者：赵林静 ZHAO Lin-jing(School of Computer,Civil Aviation Flight University of China,Guanghan 618307,China)

机构地区：[1]中国民航飞行学院计算机学院

出　　处：《计算机工程与设计》2019年第12期3514-3519,共6页Computer Engineering and Design

基　　金：国家自然科学基金民航联合基金重点项目(U1233202/F01)

摘　　要：为对评论文本进行准确的主题分类,提出一种结合HowNet语义相似度和隐含狄利克雷分配(LDA)模型的主题聚类方法。不同于传统LDA模型,该方法通过HowNet常识知识库计算输入单词与当前主题聚类中单词间的语义相似度,以此调整LDA模型中的超参数β。为不同的单词分配不同的β值,以此监督聚类过程,在主题分析中实现从语法到语义的转变。实验结果表明,该方法能够有效提高主题聚类的准确性。To classify comment texts accurately,a topic clustering method based on HowNet semantic similarity and implicit Dirichlet distribution(LDA)model was proposed.Different from the traditional LDA model,the semantic similarity between the input words and the words in the current topic clustering was calculated through HowNet common sense knowledge base,and the hyper-parameterβin LDA model was adjusted.Differentβvalues were assigned to different words to supervise the clustering process,thus realizing the transformation from grammar to semantics in thematic analysis.Experimental results show that the proposed method can effectively improve the accuracy of topic clustering.

关 键 词：评论短文本 主题分析 HowNet语义相似度 LDA模型 超参数β 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]