Research On Prediction Method Of Scientific Research Cooperation Based On Network Embedding

With the development of society and the advancement of technology,more and more scientific researchers are involved in academic research.In recent years,many researchers have used the results of academic papers as the research background,exploring cooperative prediction problems among scientific researchers through link prediction methods.Based on the analysis of current mainstream link prediction methods,this paper found some shortcomings in above method.In order to carry out link prediction in scientific research cooperation network accurately and efficiently,this paper will take the network embedding into the link prediction of research cooperation network,and propose a research cooperation network embedding algorithm based on the random walk of author’s effect.This paper first considers the author’s previous research cooperation relationship and calculates the effect of the author node in the research cooperation network.Then,sampling the author node context according to the node effect.With the help of context sampling results,the author node is learned by SkipGram model in Word2vec and the author node representation vector is obtained.Finally,Evaluation of author similarity by the angle between representation vectors calculated by cosine similarity function,and completes the scientific research cooperation prediction task.The main contributions of the thesis are as follows:(1)According to extract scientific research cooperation relationship to construct scientific research cooperation network on the DBLP dataset and the Topic-paper-author dataset of the Aniner platform.taking the effect of the author node based on calculating the time factor and the network topology structure as the reference of the random walk process.Sampling the context of the author node,and the network structure around the author node is obtained.(2)With the help of the context sampled,the author nodes in the scientific research cooperation network are represented by SkipGram model,and the author nodes’ representation vectors are obtained.The representation vector of the author node is used as the input of the cosine similarity function,and calculating the cosine similarity distance between the authors,then taking the result as the prediction of the author’s basis for cooperation to achieve the prediction of scientific research partners.(3)Cooperative prediction experiments of scientific research are carried out on the data sets.The experimental results show that,compared with the classical similarity index method,the research cooperation prediction method based on network embedding proposed in this paper has better prediction effect,and has better prediction effect Compared with the algorithm of classification learning process,the prediction method proposed in this paper has higher execution efficiency and can accomplish the prediction task of large-scale network efficiently.