Abstract: Most current expert retrieval methods based on scientific literature get the expert information statically. Meanwhile, the dynamic evolution of analytical methods are seldom applied in expert retrieval, with little regard of external information such as literature authors, citations authors. Based on this, Citations Author Topic over Time（CAToT） model is constructed on the basis of the CAT model and ToT model. And a Gibbs sampling method to estimate model parameters of CAToT is given, the same as the methods applied in the expert retrieval. The model which gathers the advantages of CAT model and ToT model can not only reveal the hidden topics and the related authors and citations, but also the rules of topics and expert ranking evolution change over time. Finally, the extensive experimental results from 1557 articles from ACL, CONLL, EMNLP conference papers indicate that AToT model is feasible and efficient through comparative analysis with the CAT model.

Key words: experts retrieval, citation theme evolution model, Gibbs sampling, scientific literature

摘要： 目前基于科技文献的专家检索方法大多数是静态地获取专家信息，而动态演化的分析方法很少考虑文献的作者、引文作者等外部信息，且很少应用于专家检索领域。基于此，在CAT和ToT模型的基础上构建了引文作者主题演化（CAToT）模型，并给出了一种估计CAToT模型参数的吉布斯采样方法以及该模型在专家检索方面应用的方法。该模型集成了CAT和ToT模型的优势，不仅可以揭示科技文献中隐含的主题、与主题相关的作者和引文作者，而且可以挖掘主题随时间变化的规律以及专家排名的演化规律。以1 557篇ACL、CONLL、EMNLP的会议论文集作为实验数据，通过与CAT模型的对比分析验证了CAToT模型的可行性和有效性。

关键词: 专家检索, 引文主题演化模型, 吉布斯采样, 科技文献