Research On Prediction And Evaluation Of Paper’s Influence Based On Academic Network

To realize reasonable and effective evaluation of the influence of academic papers is of nice significance for exciting the innovation potential of researchers,creating the innovation environment and even promoting the establishment and development of the national science and technology innovation system.Predicting and evaluating the academic impact of papers based on network structure is a hot topic in recent years.However,most of the current prediction work focuses on the prediction of the overall citation increment of the network structure,but lacks the prediction of the future citation trend of a single paper,which makes it difficult to directly apply the prediction results to the impact evaluation of a single paper.In the research of academic impact evaluation based on network structure,most of the work is based on Page Rank algorithm to evaluate the impact of papers in the whole network through the way of impact iterative calculation.However,these works fail to distinguish the difference of influence transmission intensity caused by different citation behaviors,so it is difficult to obtain effective evaluation results.In this paper,the prediction and evaluation of the influence of academic papers based on network structure are deeply discussed,and the strategies to deal with the above problems are proposed respectively.The main research contents of this paper are as follows:(1)Research on incremental citations prediction of a single paper based on the citations cascade structure.Based on the citations cascade structure formed by the citations relationship between papers,this paper proposes a model ICPCI to predict the future citations increment of a single paper.The ICPCI model constructs the citations cascade structure according to the citation relationships of the target papers within one year after publication,which is used to represent the hierarchical structure and time characteristics of citations between papers.Based on the reference cascade structure,the initial representation vector of the paper is generated by hierarchical aggregation.The initial paper vector is input into the bidirectional gated cycle unit(Bi GRU)network,and the paper features are extracted in order and input into the fully connected layer to realize the prediction of the future citation increment of a single paper.The experimental results on the public data set show that compared with the baseline model,the ICPCI model can achieve better reference increment prediction performance.(2)Research on academic impact evaluation of papers based on heterogeneous academic networks.Based on the heterogeneous academic network composed of different academic entities,this paper proposes a new algorithm to evaluate the academic impact of papers(AIRank).The contribution of AIRank algorithm is mainly reflected in the following two aspects: a)Aiming at the limitations of the current evaluation algorithm in the average distribution of influence between node pairs,AIRank algorithm quantified the intensity characteristics of influence transfer between nodes on the premise of examining the emotional attribute of reference between node pairs,semantic similarity and academic quality difference,and thus realized the reasonable distribution and transfer of influence between different node pairs;b)Combined with the mutually reinforcing relationship of influence among heterogeneous academic entities,AIRank algorithm fine-tuned the academic influence of heterogeneous neighbors on academic paper nodes to evaluate the academic impact of papers more comprehensively.The experimental results show that compared with Page Rank algorithm,AIRank algorithm can distinguish the academic influence of papers more accurately and reasonably.