Simulation Of Pure Citation Network

Citation,as one of the most important tool of scientometrics,has been widely used in evaluation of academic units like scholars,institutions,and journals.The topics related to citation behavior and citation impact analysis become more and more popular.The methods used in these researches switch from correlation evaluation methods like regression analysis to causal inference recently.However,it is controlversial to use the historical data to infer causal relations between impact factors and citation.Since citation process are influenced by these impact factors in dynamic ways which have been lost in such kinds of static data like citation network.Actually,to use such kinds of historical data is a comprising situation since we really do not have more accurate data.In this paper,we attempt to propose a method to simulate the formation process of citation network,so as to provide the lasting data for researches.Based on these lasting data,impact factors could be putted in simulation process as simulus,and causal relations could be infered through the comparison between different simualtion results.Based on the above idea,we propose a single topic pure citation network model in chapter 3 which assume that authors in a domain are worked for one topic and articles are cited based on their value.The value of an article is defined as the knowledge gain of this domain after publishing this article.The more value an article has,the higher probability being cited.Science is a process of knowledge accumulation,knowledge gain of an article is an iterative process based on present knowledge.Based on this idea,this paper propose an iterative value generation method which generate value of an aritcle according to its average value of reference defined as reference value.The value coefficent which is the ratio of article value and reference value follows log-normal distribution by learning from the real data.A pure citation network could be generated through simulation by using the iterative value generation method and displines such as author increment,author productivity learned rom data.In chapter 4,a multi-topic pure citation network is proposed.In this model,we assume the probability an aritcle cited by articles with topic A is subject to the value it brings to topic A.Value of an article changes from an scalar number to an vector in which value to every topic are stored.The reading capacity of authors can not space with the incresement of articles.In chapter 5,we propose three reading rule,based on which authors build their own personal article repository.Only artiles in personal repository can be selectd.Based on single topic and multi-topic pure citation models,simulation are implemented.In chapter 6,we attempt to explain the relation between value coefficent distribution and citation distribution.By using value coefficent distributions with different parameters,the final citation distribution of simulation result could follow power law distribution,exponential distribution and log-normal distribution.Two application are implemented in chapter 7 and chapter 8.Based on pure citation network model,factors related to literature aging and scholary impact facotrs are putted in the process of simulation,causal inference could be made through comparing results.