基于OLDA的可变在线主题演化模型
发布时间：2018-08-09 12:37
【摘要】：【目的/意义】随着网络社交媒体的发展,舆情文本中隐含的主题越来越能体现出人们的关注点所在及变化情况,因此对其进行检测及演化分析具有重要意义。【方法/过程】为了解决OLDA模型存在的主题混合及权重定义问题,本文提出了一种可变在线LDA模型(variable online LDA,VOLDA),通过构建主题相似度矩阵,明确主题变化关系,在主题内容演化矩阵中剔除含有旧主题的时间片,从而构建变长的演化矩阵,并在此基础上设计动态权重计算方法及先验参数优化方法。【结果/结论】基于论坛文本数据的实验结果表明,VOLDA模型能够有效减少新主题出现后的主题混合问题,并且提高主题在演化过程中的表示能力。
[Abstract]:[purpose / meaning] with the development of online social media, the topics implied in the text of public opinion increasingly reflect people's concerns and changes. Therefore, it is of great significance to detect and analyze its evolution. [method / process] in order to solve the problem of topic mixing and weight definition in OLDA model, In this paper, a variable online LDA model, (variable online LDA-VOLDA), is proposed. By constructing the topic similarity matrix, the relationship of topic variation is clarified, and the time slice containing the old theme is eliminated in the topic content evolution matrix, thus the variable length evolution matrix is constructed. On this basis, a dynamic weight calculation method and a priori parameter optimization method are designed. [results / conclusions] the experimental results based on the forum text data show that the VOLDA model can effectively reduce the topic mixing problem after the new topic appears. And improve the expression of topics in the evolution process.
【作者单位】： 南京航空航天大学经济与管理学院;
【基金】：国家自然科学基金项目(71373123) 江苏高校哲学社会科学研究重点项目(2015ZDIXM007)
【分类号】：TP391.1
，
本文编号：2174096
[Abstract]:[purpose / meaning] with the development of online social media, the topics implied in the text of public opinion increasingly reflect people's concerns and changes. Therefore, it is of great significance to detect and analyze its evolution. [method / process] in order to solve the problem of topic mixing and weight definition in OLDA model, In this paper, a variable online LDA model, (variable online LDA-VOLDA), is proposed. By constructing the topic similarity matrix, the relationship of topic variation is clarified, and the time slice containing the old theme is eliminated in the topic content evolution matrix, thus the variable length evolution matrix is constructed. On this basis, a dynamic weight calculation method and a priori parameter optimization method are designed. [results / conclusions] the experimental results based on the forum text data show that the VOLDA model can effectively reduce the topic mixing problem after the new topic appears. And improve the expression of topics in the evolution process.
【作者单位】： 南京航空航天大学经济与管理学院;
【基金】：国家自然科学基金项目(71373123) 江苏高校哲学社会科学研究重点项目(2015ZDIXM007)
【分类号】：TP391.1
，
本文编号：2174096
本文链接：https://www.wllwen.com/kejilunwen/ruanjiangongchenglunwen/2174096.html
