CAJ | 학술논문

本文以“科学知识网络中的链路预测”为主要对象，对链路预测的类型、研究思路和方法等相关理论进行了回顾，将知识网络划分为同质网络和异质网络两种类型，从合作网络、引证网络和二分网络三个方面对同质网络的研究进行梳理，并介绍了一些异质网络中的链路预测方法。认为：针对这方面的研究近年来有成为图书情报学领域研究热点的趋势；已有研究多是描述各种链路预测指标在不同类型知识网络中的预测效果，未来应当利用链路预测量化和评价演化模型，识别和分析异常链路，以发现知识热点和创新趋势，将知识网络的研究提升到应用层次。图5。表2。参考文献68。
본문이“과학지식망락중적련로예측”위주요대상，대련로예측적류형、연구사로화방법등상관이론진행료회고，장지식망락화분위동질망락화이질망락량충류형，종합작망락、인증망락화이분망락삼개방면대동질망락적연구진행소리，병개소료일사이질망락중적련로예측방법。인위：침대저방면적연구근년래유성위도서정보학영역연구열점적추세；이유연구다시묘술각충련로예측지표재불동류형지식망락중적예측효과，미래응당이용련로예측양화화평개연화모형，식별화분석이상련로，이발현지식열점화창신추세，장지식망락적연구제승도응용층차。도5。표2。삼고문헌68。
Currently, the research about link prediction of knowledge network is scattered in the fields of Statistical Physics, Computer Science, Complex Networks and Library and Information Science. From the perspective of Library and Information Science, this paper mainly reviews researches on link prediction of knowledge network and systematically analyzes previous researches. This paper used ＂link prediction＂ as topic to retrieve the literatures in WoS, and got 300 records. Using CitNetExplorer to analyze the direct citation relationships among the 300 records, we obtain the highly cited interrelated literatures to review on current link prediction types and research ideas. Link prediction can be classified into two types： the static and the dynamic, and the corresponding dataset partition methods are different. The former uses random sampling, while the latter needs to consider the temporal state. In the field of Library and Information Science, knowledge networks vary in size and scale. So the link prediction of such knowledge networks uses the similarity-based algorithms in order to reduce computing complexity, but also introduce certain semantic and attribute information to ensure the accuracy of prediction. This paper divides the knowledge network into homogeneous and heterogeneous networks. For the link prediction of homogeneous network, this paper reviews on the research progress from the aspects of co-au- thorship network, citation network and bipartite network. The co-authorship network can be viewed as an undirected network, and is the easiest way to describe the real network system. This paper summarizes the predictors and steps in link prediction of co-authorship network and examines its prediction effect from au- thor, institution, and country level. The citation networks can be viewed as a directed network, and is the first proposed knowledge network. Compared to the co-authorship network, the citation network not only has the structure information of basic data, but also involves the external information, such as authors, journals and content of articles. Therefore, the citation network not only can be predicted based on the local struc- ture information, but also by the external information, or by the integration of structural and external infor- mation to perform machine learning to improve the accuracy of link prediction. The bipartite network is a special kind of homogeneous network, which can be used to observe structural characteristics and evolution process between subject and object. Using link prediction in bipartite network can solve some problems of recommendation systems. For the link prediction of heterogeneous network, this paper reviews on some link prediction methods, such as forecasting model based on meta-path. This paper finds out that link prediction of knowledge network in Library and Information Science has be- comes a hot issue in recent years. Most of current studies are empirical studies describing the prediction effect in different types of knowledge networks based on a variety of predictors in order to determine the ap- plication scope of link prediction. This paper proposes that in the future research, the link prediction should be used to quantify and evaluate the evolution models, to identify and analyze anomalous links, and to discover knowledge hot spots and innovation trends. Ten years of development shows that knowledge net- work can be used as an ideal cartier for link prediction, and link prediction is a powerful tool for analyzing knowledge network. On this basis, carrying out applied research from link prediction perspective based on the structure and evolution of knowledge network will be the future research direction of knowledge network in the field of Library and Information Science. 5 figs. 2 tabs. 68 refs.