Research On Knowledge Graph-based Recommendation Approach For Scientific Research Project Collaborators

As scientific research continues to grow,the declaration and implementation of research projects become more complex,and quality research collaboration can provide strong support for the advancement of projects.It is a challenge for researchers to find suitable collaborators for research projects efficiently and accurately.The traditional method of collaborator recommendation in the research field has the problem of not fully utilizing research information,such as researcher information and collaboration relationships,which limits the effectiveness of recommendation.Since the knowledge graph can adequately represent the semantic connections between entities and deep learning has excellent knowledge representation capability,this paper uses deep learning combined with the knowledge graph to focus on the construction of knowledge graphs in research fields and proposes a representation learning method applicable to this knowledge graph and a recommendation algorithm for research project collaborators based on the knowledge graph.The main work of this paper is as follows:(1)Analyze the data characteristics of the research field,collect the research data related to the research projects,and finalize the construction of the knowledge graph pertaining to the research domain.(2)Researched the learning method of knowledge map representation,and aimed at the problems of insufficient diversity of positive samples and large randomness of negative samples in the sampling of traditional translation models,the Enhanced Trans D model was proposed,and the sampling method and loss function of the Trans D model were improved.The collaborators recommend that accuracy provide the premise.(3)The Research Project Collaborator Recommendation Algorithm Based on Knowledge Graph(RPCR-KG)algorithm is proposed,which uses the combination of knowledge graph and deep learning technology.Based on the information of scientific researchers,the relevant information of scientific research cooperation in the knowledge map was integrated.The algorithm achieves recommendation tasks through steps such as subgraph recall,scientific project ranking,and scientific project collaborator ranking.(4)Based on the algorithm proposed in this paper,a scientific research project collaborator recommendation system is built to design and implement the architecture and functions of the system,complete the testing of the system,and verify the usability of the system.The research results show that the Enhanced Trans D model proposed in this paper improves the accuracy of the triplet classification task by 3.2%-10.94% compared with the traditional translation model,enhances the diversity of sample sampling,reduces the randomness of negative sample sampling,and improves The model represents the performance of the knowledge map;the RPCR-KG algorithm introduces the knowledge map into the recommendation of scientific research project collaborators,so that the information of researchers and the network information in the knowledge map can be effectively combined,and it is more accurate than the optimal baseline model on the same test set.The rate and average reciprocal ranking increased by 4.19% and 1.85%respectively,improving the accuracy and quality of the recommendation;the scientific research project collaborator recommendation system can make more accurate recommendations,providing a reference for researchers to find suitable scientific research project collaborators.