Research On Influence Maximization Of Citation Network Based On Meme

With the rapid development of information technology and the vigorous promotion of knowledge economy,scientific research has entered the era of big data.In recent years,a large number of publications and citation data known as "meta-knowledge",as well as leaps in the theory and modeling of complex systems,have promoted the large-scale exploration of science.Citation networks,which are composed of mutual citation relations of scientific literature,have attracted more and more attention of scholars and become an important medium of scientific research.The problem of influence maximizing of citation network focuses on the most influential scientific literature and makes them reach the maximum influence through spread,which is of great practical significance for promoting the development and progress of science.However,traditional research is often carried out based on network topology,ignoring the huge value of semantic information contained by nodes in citation networks,which results in the quality of selected seed nodes cannot be guaranteed,and it is difficult to obtain satisfactory research results with influence maximization.Accordingly,how to scientifically model the node information transmission mode and combine the existing citation network to design an effective influence maximization method has become one of the current research hotspots.Memes are short text units of scientific literature,which widely exist in complex networks and play an important role in analyzing citation content and communication mode.Based on the topological structure of node interaction,the classical information transmission model spreads influence by randomly selecting propagation rate or node affected threshold,and fails to give a quantitative expression from the perspective of content transmission.By applying the meme to the network influence spread of citation,we can explore the mode of knowledge diffusion in finer granularity and quantify the content transmission of scientific literature in micro level.Therefore,introducing memes to carry out research on influence maximization of citation networks is more consistent with the transmission characteristics of real citation networks,and the results are more realistic.The main research works of this thesis are as follows:(1)The influence propagation model of citation network based on meme basic unit is studied.Based on the analysis of the advantages and disadvantages of the current influence propagation model,aiming at the neglect of node semantic information in two classical information propagation models(independent cascade model and linear threshold model)and the randomness in the process of influence propagation,the meme analysis method is introduced to study the selection of inter node propagation rate and node affected threshold.The calculation methods of quantifying the influence between nodes with edge propagation score and quantifying the affected threshold of nodes with node propagation score are proposed,and then two improved meme cascade model and meme threshold model are obtained.Finally,the effectiveness of the models are verified by propagation experiments.(2)Based on the two influence propagation models,aiming at the information redundancy in the process of mining the most influential node set in directed unauthorized networks,a discount strategy based on Leader Rank is proposed to effectively alleviate the influence overlap effect between nodes,and then the seed nodes are selected by greedy algorithm.Research and design a combination of discount strategy and greedy thinking influence maximization method.(3)The proposed method is verified by experiments on five real citation network data sets in different fields.The data source is preprocessed to obtain the citation information data set and reference relationship data set respectively,and the meme extraction method is used to identify the meme in each network.In order to verify the effectiveness and efficiency of the proposed method,the differences among different methods are compared from the perspectives of propagation dynamics and time efficiency.The experimental results show that the proposed method is better than other heuristic methods.In terms of influence extension,its influence spread after spreading through the meme cascade model is at most 2.3 times that of the degree centrality method,2.5 times that of the betweenness centrality method,and its influence spread after spreading through the meme threshold model is at most 17 times that of the degree centrality method,and 5.7 times that of the betweenness centrality method.Furthermore,the proposed method can reach the influence spread close to CELF in both models,and the time efficiency is higher than CELF,which is suitable for large-scale citation networks with sparse edges.