Scientific Research Team Identification And Service Components Based On Federated Learning

Multi-type,multi-morphological and ubiquitously connected scientific and technological big data contain a large amount of data information related to scientific research teams,and constitute a complex,large-scale,and interconnected academic heterogeneous information network.Most of the existing studies are only applicable to homogeneous information networks,ignoring the rich semantic features and topological features contained in the academic heterogeneous information networks composed of scientific research teams.In terms of federated learning,existing research team identification methods cannot solve the negative impact of federated working nodes on parameter aggregation updates due to differences in the local computing capabilities,which brings great difficulties and challenges to the identification of scientific research teams in scientific and technological big data.Due to the increasing specialization,complexity and difficulty of scientific research tasks,the cooperation of scientific research teams has also become an inevitable trend to promote the development and progress of science and technology.How to effectively identify and discover scientific research teams from massive and complex scientific and technological big data has become an urgent scientific problem to be solved in the field of scientific research.Scientific research team identification and service components based on federated learning has practical significance for helping researchers quickly grasp the status quo of the team,helping team leaders improve team management and guiding the team to develop better.The main contributions of this thesis are as follows:(1)A federated learning-based academic heterogeneous information network representation learning method is proposed,which realizes the low-dimensional and dense real-valued vector representation of nodes under the premise of preserving the structural and semantic features of nodes in the network.The topological and semantic features of nodes in academic heterogeneous information networks are deeply mined through node-level and meta-path-level attention mechanisms.By combining the representation learning of academic heterogeneous information network with federated learning and applying the version control of parameter weights and version threshold detection,the dynamic aggregation update of federated parameter weights is realized,which reduces the negative impact of federated working nodes on parameter aggregation updates due to differences in the local computing capabilities.(2)A method for evaluating the importance of node in academic heterogeneous information network based on network node-level attention weight coefficients is proposed.Based on the measure of similarity between nodes,mine the similarity relationship of nodes in academic heterogeneous information network in terms of semantic and topological features.And the node-level attention weight coefficients are used to measure the influence of neighbor nodes and finally realize the importance evaluation of nodes in academic heterogeneous information composed of scientific research teams by aggregating the influence of neighbor nodes.(3)A scientific research team identification method based on the maximization of node influence is proposed.Based on the evaluation results of node importance,the influence weight coefficient of node-level attention is used to sort the influence of node neighbors.Through he maximization of node influence,the effective identification of the "leader"of the scientific research team is realized.Based on the analysis of similarity relationship between nodes and the "leader" node,and the analysis of the influence relationship between neighbor nodes,effective identification and discovery of the core and non-core members in scientific research teams is realized.(4)A scientific research team identification and service component based on federated learning is designed and implemented,which integrates the methods and algorithms proposed in this paper,and provides users with scientific research team identification and discovery functions,including basic information query and retrieval of scientific research teams,detailed information query and retrieval of scientific research teams,and detailed information query and retrieval of team members.And we designed an intuitive and concise visual display interface for the identification and discovery results of scientific research team.