基于层次概率主题模型的科技文献主题发现及演化

Topic Extraction and Evolution for Scientific Literature Based on Hierarchical Probabilistic Topic Model

自动挖掘科技文献主题并识别主题变化对于科研工作者及时获取相关领域的最新研究动态有着重要作用。针对科技文献主题多样、动态性强等特点,分析科技文献主题发现及演化具体方法,基于层次概率主题模型h LDA,采用Gibbs抽样来进行模型参数估计,并运用互信息的方法对主题词进行筛选,以提取高质量的主题词。最后,利用先/后离散分析方法研究主题随时间的演化问题。实验结果验证了主题发现及演化方法的可行性及有效性。

Automatic mining scientific literature＇ s topic and observing topic change for researchers will play great role in understanding and accessing the latest research frontiers on certain field. This paper analyzed topic extraction and evolution approaches of scientific papers by examining the characteristics of the diversity and dynamics of scientific papers, and based on hierarchical probabilistic topic model, using Gibbs sampling to estimate the model parameters and choosing the high - quality topic words by means of mutual information. This paper finally used Pro/Post - discretized analysis to study the topic evolution. The experimental results show that topic extraction and evolution method proposed in this paper are feasible and effective.