Research On Scientific Research Relationship Based On Complex Network

At present,researchers continue to produce scientific research results in scientific exploration and publish them to the society,thus producing a large number of scientific research data.The development of data makes the retrieve,management and analysis of scientific research data more and more challenging.Therefore,how to accurately retrieve the required data from the huge scientific research data,and extract effective information from the data for analysis,has become a concern that researchers pay close attention to.This thesis explores the corresponding relationship between the author of the same name and the paper in scientific research data,as well as the cooperative relationship between authors.Firstly,aiming at the problem of homonymy ambiguity in the correspondence between the author's signature and the author's entity in real life,this proposes a homonymy disambiguation algorithm based on heterogeneous network and clustering.This first transforms the homonymous disambiguation research under the background of scientific research data into the clustering research of homonymous papers,designs the semantic similarity and discrete feature similarity between papers,and takes these two similarities as the basis of paper clustering.Subsequently,in order to characterize the relationship of discrete feature between papers,the network representation learning method based on random walk of meta path in heterogeneous network is adopted.Furthermore,the contribution ratio of semantic similarity and discrete feature similarity to the final paper similarity is determined by experiments,as well as the combination of edge types used in constructing heterogeneous networks.Finally,the experimental results demonstrate the effectiveness of each step of the homonymy disambiguation algorithm and the effectiveness of the whole algorithm.Furthermore,this thesis improves the label propagation algorithm in community discovery to obtain close circle and mine the potential cooperation between authors.Firstly,in order to design effective attributes,this analyzes the characteristics of a specific network of research collaboration.Next,the problems existing in the traditional label propagation algorithm are analyzed,and the node update order and label selection rules are improved.Then,this proposes a label propagation algorithm which combines multiple neighborhood overlap rate and historical label similarity(NOHLPA).Furthermore,the parameters used in NOHLPA are determined by experiments.Finally,the experimental results demonstrate the effectiveness of each step of NOHLPA and the effectiveness of the whole algorithm.Then,NOHLPA is used to divide the specific scientific research cooperation network,which proves that NOHLPA can obtain close and accurate cooperation circle.