Research On Recommendation Algorithm Of Cross-domain Scientific Collaborator Based On Big Scholarly Data

In recent years,cross-domain scientific collaboration has a positive impact on scientific technological development and academic innovation.However,with the increasing number of researchers and the formation of big scholarly data,it is becoming more and more difficult and time-consuming for researchers to find collaborators outside their professional domains.Due to the differences between different domains,most researchers are not familiar with other domains,and it is a great challenge to correctly identify research sites and collaborators in other domains.Therefore,this paper studies how to automatically recommend appropriate cross-domain collaborators for researchers based on big scholarly data.At present,some scholars have conducted researches on the recommendation of cross-domain research partners,but these researches only focus on the cross-domain collaboration topic discovery,and do not consider whether two researchers from different domains have the same academic level.In fact,in the real life,people prefer to choose a partner who has the same social status as himself.Therefore,we assume that the collaborators who have the similar academic level will do better during the collaboration.based on this assumption,we try to design a new kind of cross-domain scientific collaborators recommendation algorithm,this algorithm not only consider the correlation research topic of the researchers,at the same time also to consider their academic level of similarity.The first question we need to solve is how to determine whether two researchers from different domains are of similar academic level.It is difficult to use a unified standard to measure all domains due to the differences in the measurement of academic level between different disciplines and domains.Therefore,this paper introduces clustering algorithm to group researchers in each domain.We made a comprehensive analysis and summary of the evaluation indicators and methods of researchers' academic level,and finally selected five indicators,namely academic age,total number of papers published,total number of citations,number of citations per article,and number of collaborators,as the basis for the grouping of researchers.The mini batch k-means algorithm was used to group researchers according to their academic level.Then the groups of each domain are matched one by one to make the researchers in thecorresponding groups have the same academic level.Finally,the target scholar is recommended to the cross-domain research partner with similar academic level and the most relevant research interests.We obtained the paper data of bioinformatics,data mining and molecular biology from Microsoft Academic Graph,and obtained the experimental data set after processing.The algorithm proposed in this paper is verified on this data set,and the algorithm has good performance in three aspects of recommendation precision,recall and F1 score.