Academic Team Recognition And Team Member Recommendation Based On Network Motifs

Science of Scientific Team Science is one of the important research directions of the Science of Science.The effective recognition of academic teams is the basis for analyzing and understanding the collaboration pattern of academic teams.The recommendation of academic team members is an important strategy to optimize the academic team.Researches have shown that academic teams with higher familiarity among members will achieve higher teamwork efficiency and team performance.However,existing studies lack a measure of familiarity between scholars and other team members.Therefore,based on the network motif,this paper proposes multi-level team familiarity and proposes an academic team recognition method and an academic team member recommendation method that combine team familiarity.First,we analyze the research background and significance of academic team-related problems and team familiarity,and summarize solutions to academic team recognition and member recommendation problems in related works.Then we introduce the concept of network motifs,and summarizes the structural and statistical measures for network motifs.Then based on the network motifs,the concept of team multi-level familiarity are put forward.The familiarity is applied to academic team recognition and member recommendation.For the problem of academic team recognition,we propose an academic team recognition algorithm based on the multi-level familiarity of the team.This paper performs preliminary clustering based on the closeness of the scholars' collaborative relationship,and optimizes it based on team familiarity and node local density.By extracting the paper information of the MAG data set,a data set containing the author's organization information is integrated.The team identification algorithm proposed in this paper is applied to the data set to verify the effectiveness of the method.Through the information analysis of the identified academic team and academic institutions,it is found that the proportion of cross institutional teams increases year by year and the performance is higher under a certain team scale.For the problem of recommendation of academic team members,this paper proposes a graph kernel similarity calculation method,which considers the scholars' skill matching,structure matching and team familiarity improvement,and recommends the candidate with the highest similarity for the team.By analyzing the paper information of the CiteSeerX data set,the experimental set and verification set containing the co-authored team are obtained.Experimental results show that the algorithm proposed in this paper has a higher accuracy rate and is better than the benchmark method in terms of team output quality,communication cost and team structure.