科学在人类社会中扮演着越来越重要的角色，研究科学家的引用行为可以让人们对科学的发展和知识的传播有更深入的认识。然而以往的研究主要集中在引文网络的拓扑结构性质，忽略了文章的内容信息，所以本文希望在引文网络中引入文章内容，进一步研究学者的引用模式。 本文采用LDA模型训练APS数据库中的文章摘要信息，从中提取出高维主题向量来表示文章内容，并映射到高维学术空间中，用向量之间的距离表示文章之间内容的差异。将文章学者分为三个类别：自己、明星学者和普通学者，研究学者的引用行为与文章距离与引文学者类型之间的关系。学者的引用行为具有集中性和右偏性。当文章距离在[0,0.1)范围内，引用偏好随着文章距离的增大而增大；在[0.1,0.5)内，引用偏好随着文章距离的增大而减小，且服从指数分布；在[0.5,1)内，引用行为较为随机，随距离变化不明显。学者自引的偏好要明显高于引用明星学者的和引用普通学者的偏好，引用明星学者的偏好也略高于引用普通学者的偏好。之后，本文建立了一个引用的心理力模型模拟学者对引文的选择，得到了很好的效果。

Science plays an important role in our society. Researching on the citation behaviors of scientists can let people know more about the development and spread rules of science. However, the previous researches mostly focused on the topological structure of citation networks but overlooked the contents of papers. So this paper introduced the contents of papers into citation networks and researches scientists’ citation behaviors under different distances. This paper uses LDA model to analyze the abstracts of papers from APS dataset and then extracts the high-dimension topic vector to represent the contents of papers. It measures the distance between two topic vectors as the diversity of papers. The authors are divided into three categories: self, star and others. This paper finds that the citation of physicists is concentrated and right-skewed. When the distance between two papers is under [0, 0.1), the citing preference increases as the distance increases. When the distance between two papers is under [0.1, 0.5), the citing preference decreases as the distance increases and obeys the power-law. When it is under [0.5, 1), the citing preference is almost random. The preference of self-citation is obviously higher than that of star-citation and that of other-citation. The preference of star-citation is slightly higher than that of other-citation. What’s more, this paper builds an multi-agent citation model to simulate the citation behaviors of scientists and the model fits the reality well.