【作者】 张超；

【导师】 刘怀兰；

【作者基本信息】 华中科技大学 ， 机械电子工程， 2018， 硕士

【摘要】 随着移动互联网、云计算、区块链等新兴领域的飞速发展,人们的生活方式随之发生了深刻的变化。为了能够预判领域发展,并采取有效应对措施抢占先机,相关人员需要对领域的发展历程有着充足的认知和理解。因此如何快速识别某个特定领域的演化路径成为亟待解决的问题,针对这一问题,本文开展了如下工作:文章首先调研了先前领域演化路径识别方法的特点,发现传统方法存在着诸多局限性,难以适应文献爆发性增长的现状,利用主题变迁可以有效应对这一问题,并且能够挖掘出更深入的主题信息。接着提出一种基于主题变迁的领域演化路径识别方法。该模型可以自动从Aminer平台获取数据,通过挖掘科技文献中隐藏的语义信息,得到不同时间段的研究主题;利用Jaccard相似度计算不同主题之间的关联,得到演化路径并进行可视化展示。通过对人工智能领域的实证分析,结果表明该模型能够有效反映领域研究主题的变迁。最后针对领域主题挖掘进行了深入的研究。为了解决传统主题模型中主题数量需要预定义、缺乏外部信息的问题,提出了融合引文信息的层次狄利克雷过程(ciHDP)。为了验证算法的有效性,本文使用Cora和CiteseerX数据进行了定量的对比实验,同时使用Aminer数据进行了定性的分析,定量和定性的结果共同说明了改进后的方法在主题挖掘分析方面的有效性。更多还原

【Abstract】 With the rapid development of emerging areas such as mobile internet,cloud computing and blockchain,people’s lifestyles have undergone profound changes.In order to be able to predict the development of the field and take effective countermeasures to seize the opportunity,relevant politicians,business people,and scholars need to have sufficient knowledge and understanding of the development process of the field.Therefore,it is urgent to quickly identify evolutionary paths in a specific field.To solve the problem,this article has carried out the following work:The article first investigated the characteristics of the evolution path identification method in the previous field,and found that the traditional method has many limitations,making it difficult to adapt to the current situation of the explosive growth of the literature.This article next attempts to use text data mining methods for research.However,most of the current thematic change methods cannot simultaneously consider the changes of the subject content and the changes of the subject-related scholars.Therefore,this paper proposes new methods and processes to solve this problem.Then it proposes a domain evolution path identification method based on topic transition.The model can automatically obtain data from Aminer platform.Through mining the semantic information hidden in scientific literature,the research topics in different time periods can be obtained.The Jaccard similarity is used to calculate the correlation between different topics,to obtain evolutionary paths and to visualize them.Through empirical analysis of the field of artificial intelligence,the results show that the model can effectively reflect the changes in the field of research topics.Finally,we conducted in-depth research on domain topic mining.In order to solve the problem that the number of topics in the traditional topic model needs to be predefined and lack of external information,a hierarchical Dirichlet process(ciHDP)for merging citation information is proposed.In order to verify the validity of the algorithm,the author used the data of Cora and Krypton as the experimental data set to carry out quantitative comparative experiments.At the same time,qualitative analysis was performed using Aminer data.The results of quantitative and qualitative analysis jointly demonstrated the improved method in topic mining.Analytical validity.更多还原