Research On The Recommendation Of Scholars Based On User Portraits And Cooperative Relationships In Scientific Research Social Networks

With the development and popularization of Web 2.0 technology,scientific research social networks attract a large number of users.Users share research results,participate in topic discussions,and create research groups on the network,which effectively improves the efficiency of scientific research.The academic resources in scientific research social networks are increasing rapidly,and there is a problem of "information overload".Information overload limits the information search capabilities of scientific researchers.Therefore,how to find satisfactory collaborators in massive academic resources has become an urgent problem.The recommendation system can provide suggestions and recommended items to users,which is an effective way to solve information overload.Therefore,applying the recommendation system to scientific research social networks and establishing an effective scholar recommendation mechanism are necessary for the development of scientific research social networks.Most of the existing recommendation methods aim for paper recommendation,ignoring users' behavior seeking cooperative scholars.And the existing studies about scholar recommendation recommend scholars and research groups based on user's social relationships or scholar interests.However,providing scholars' recommendation by scholars' preferences and social relationship is few.User portraits can accurately locate scholars' preferences and analyze the interactions among scholars.The cooperative relationship reflects the strength of scholars' historical cooperation.Taking the cooperative relationship and the cooperation intensity as the side and as the weight respectively can closely link the scholars.Therefore,this paper combines the user's social information and formulates scholars' recommendation model from the researchers' user portrait and cooperative relationship.Firstly,the scientific research ability of scholars is evaluated according to the number of papers and the level of journals.Next,inputting the title,abstract and keywords of papers and using TF-IDF and LDA topic model to analyze user's performance similarity and construct user profile.This paper also analyzes user's performance similarity and constructs user profile comprehensively.Then,on the basis of considering the transmission characteristics of cooperation relationship,this paper analyzes the cooperation quality among scholars according to the cooperation relationship network,and calculates the cooperation intensity among scholars.The recommendation value of the above three dimensions is integrated to achieve the recommendation of cooperative scholars.Finally,to verify the validity of the proposed method in this paper,the comparative experiment is carried out on the real dataset.From the three aspects of precision,recall and F1,our algorithm improve 54.2%,55.5% and 54.5% respectively compared with user-based collaborative filtering,improve 89.73%,84.61% and 84.56% respectively compared with word embedding-based algorithm,improve 3.33%,26.82% and 24.02% respectively compared with social network-based recommender system within Top-10,which verify the feasibility and effectiveness of the recommended method based on user profile and cooperative relationship,and provide new ideas for the scientific researchers to find cooperative partners accurately.