Research On The Prediction Of Researchers Cooperation Potential

Scientific collaboration is one of the most important way to achieve academic results.There are many high-level results are realized by collaboration,which highlights the important position of collaboration in solving scientific problems.At present,the relevant researches on scientific collaboration are mainly in the recommendation of collaborators.Only the number of behaviors that scholars produce collaboration is studied,and the quality(level)of collaboration results is relatively rare.It is still the key that choosing the coauthors with the greatest potential to achieve the maximum benefit of collaboration.The research of collaboration potential can provide scholars with predictive guidance,which leading to select collaborators that are consistent with expectations quickly and accurately,and maximize research efficiency.With the upgrading of social life and research methods,the analysis and interpretation of cooperative behavior have become a hot topic,and the rapid development of computer technology provides a new way for the processing and mining of big data.To mine the potential patterns behind collaborative behavior based on existing collaborative big data and to predict the level of results with current collaborators is particularly critical.Some research and discussion on the prediction of the potential of the collaborators have been done in the paper following the research frontier at home and abroad.The results obtained are as follows:(1)With the explosive growth of data in the information age,it is increasingly difficult to find real and useful critical information in massive.In order to solve the problem,a prediction model of cooperative potential of researchers based on multifactor analysis using the strong learning characteristics of the ensemble learning method is proposed,which using the feature analysis and optimization,comprehensively considering the personal and related attributes of scholar's.The article title,article level,number of articles,time and order are considered to construct the sample features.The paper level of journal or conference is seemed as the sample tag of the collaborator sequence pair,indicating the potential of the current collaborators.The experimental results can converge to great values with less sample and time,indicating the superiority of the model.(2)The scientific collaboration data includes subjects such as scholars,conferences,articles,and relationships like coauthor relationships between scholars,signed relationships between scholars and articles,publishing relationships between scholars and conferences,and so on.In this paper,the nodes and edges respectively represent the subjects and the relationships to construct the Heterogeneous Information Network.So that more information is attached to the scholars who “have not published high-level articles”.Using the concept of meta path in the heterogeneous information network to predict the collaboration potential of scholars who have not published high-level articles,a Meta Path algorithm is proposed.Classification methods are used in the algorithm to predict the collaboration potential.The experimental results illustrate the feasibility of the method.In a word,based on the extraction and analysis in real scientific collaboration data,this paper quantitatively models the potential prediction problem of scientific research collaborators.The effectiveness of the algorithm is proved on real data.The research in this paper provides new methods and new ideas for the analysis of scientific research collaboration issues.There is certain value in the fields of collaborators' recommendation and social network.