| The paper citation network is a complex network formed by the citation and citation relationships between papers.It describes the results of researchers,developments in scientific fields,and relationships between disciplines.The citation relationship between papers also reveals related research content in similar fields.The paper citation network contains research results in many fields and serves as an important repository of knowledge in academic research and an important medium for scientific research.The relationship between citing papers and cited papers in the paper citation network reflects the relevance of content and the transfer of knowledge,and the citation relationship is unidirectional in time.Since the paper citation relationship is one-way and sparse,how to improve the existing paper network becomes crucial.In recent years,the total number of papers published every year has been increasing.High-quality papers can help scholars in their research in related fields.How to evaluate the impact of papers has become crucial.In order to solve the above two problems,this thesis proposes methods for link prediction and influence calculation respectively.To improve the existing paper citation network by predicting paper links,a weighted model is proposed to make the weightless paper citation network into a weighted paper citation network,the weighted model calculates the tight links on the second-order path in the paper citation network structure,the title and directed citation similarity between papers,the three calculation methods are linearly combined to give each edge a weight value;for the traditional weighted similarity algorithm,this thesis designs a damping coefficient θ,and calculates the relationship between two paper nodes with the probability of θ.The weight value accounts for the ratio of the number of links to the paper nodes and accesses the part of the common neighbor nodes between paper nodes in the weighting algorithm with a probability of 1-θ.The experimental results show that the prediction accuracy of the weighted similarity algorithm in the computer field is 0.115 higher than that of the CRA algorithm,the Common Neighbors algorithm,and the Jaccard Coefficient algorithm.The weighted model and the damping coefficient improve the accuracy of the algorithm link prediction.Aiming at the problem of paper influence calculation,this thesis proposes a ranking influence calculation method based on weighted Page Rank and HITS.In order to avoid the citation relationships of all papers in the unweighted paper citation network being treated equally,a weight calculation based on connection strength is proposed,the size of the weight value mainly depends on the number of common neighbors between the nodes of the paper;seven kinds of eigenvalues are extracted to form a feature space,and the ranking model is used for training and prediction.The experimental results show that the model accuracy of the seven eigenvalues extracted from multi-modal feature extraction in the paper citation network in the computer field reaches 0.7919.Reasonable feature extraction can improve the accuracy of the model and can effectively identify papers with high influence.In this thesis,a paper citation network visualization system is designed to visually display relevant information such as papers,authors,citation networks,institutions,etc.,and can view the newly added linked paper node information in the paper citation network diagram in the paper information details interface to find papers.Meantime,relevant papers can be searched through the influence conditions of papers. |
PDF Full Text Request