Recommendation Of Think Tank Expert Fusing Knowledge Graph Semantics

Think tanks experts play an important role in guiding and serving the development of national public opinion and other critical issues.At present,the recommendation system for think tank experts is relatively scarce,the traditional expert recommendation method based on search engine keyword retrieval has great limitations,and individuals and enterprises are eager to find experts who meet the specific needs of the industry directly and efficiently.Therefore,it is of key practical significance to study and design a recommendation method suitable for think tank experts.In order to solve the above problems,this paper studies the expert recommendation method of think tank integrating the semantics of knowledge graph.The main research contents are as follows:(1)An improved keyword extraction algorithm which is called Γ-Text Rank is proposed and apply it to the construction of knowledge graph of think tank experts.The keyword extraction algorithm proposed in this paper combines the subject influence factor of a single document with the word similarity factor of all document sets to construct the external joint influence factor,and introduces it into the word graph iterative matrix calculation process of Text Rank algorithm to complete the task of keyword extraction.The algorithm aims to expand the scope of expert domain entities to realize the construction task of expert knowledge graph of think tank experts.Comparative experiments show that the algorithm has higher keyword extraction accuracy.(2)An improved knowledge graph representation learning model which is called PTranse-SNS is proposed.The model integrates the negative similarity sampling method into the training process of the original model to improve the representation quality of entity initialization embedded vector,so as to improve the overall representation and learning effect of the model.The model can embed the graph in combination with the multi-step relationship path in the graph,which can better deal with the complex relationship path in the graph than the traditional Trans E model.Experiments show that the model gets a better representation on the expert knowledge graph of think tank constructed in this paper.(3)An expert recommendation method combining the semantics of knowledge graph and multi-dimensional expert scoring model is proposed.In this method,the entity embedding vector obtained by PTranse-SNS model is used to calculate the expert fusion similarity based on knowledge graph,and then the scoring model is constructed through the multi-dimensional information of expert attribute field.Finally,the experts are comprehensively evaluated in combination with two aspects,and the expert recommendation task is realized accordingly.Through experiments,the parameter values that make the recommendation method achieve the best performance are obtained,and an expert recommendation system is designed to verify the recommendation effect of the method in the actual recommendation pattern.The results show that the method has strong recommendation persuasion and high recommendation accuracy.