Font Size: a A A

Research On The Expert Recommendation Method Of Convolutional Neural Network Project Review Integrating Association Relationship

Posted on:2020-02-06Degree:MasterType:Thesis
Country:ChinaCandidate:X X HeFull Text:PDF
GTID:2518305969975249Subject:Control Engineering
Abstract/Summary:PDF Full Text Request
At present,information systems have been implemented in the management of science and technology projects at all levels in China,and a database of experts on science and technology projects and talents has been constructed.Experts play an important role in the evaluation of various scientific and technological projects in China.However,experts are often selected manually or randomly,which can easily lead to unmatched or unfair phenomena between experts and projects to be evaluated.Research on automatic recommendation and evaluation experts is an urgent problem to be solved in order to carry out project evaluation more efficiently.For this kind of problem,researchers usually adopt content-based recommendation method,which mainly considers the content relevance between projects and experts,and often ignores the relationship between projects and experts.Aiming at the task of expert recommendation for project evaluation,this paper proposes a method of expert recommendation for project evaluation that integrates expert Association and multi-project evaluation expert recommendation that integrates project association and expert association.The following research work has been completed:(1)Evaluation Expert Recommendation Method Fusing Expert Relevance RelationsIn the selection of experts for project evaluation,not only the discipline direction and comprehensive strength of experts themselves should be considered,but also the various social relationships among experts have important reference significance.Based on this consideration,this paper proposes a method of expert recommendation for evaluation,which integrates expert Association relations.Firstly,the correlation degree between experts is measured by several kinds of association relations among experts,and Markov network is used to represent the correlation among experts.Then,one-hot coding is used to transform the content attributes of experts and projects to be evaluated into vector representation,which is mapped into dense hidden vectors through an embedding layer,and expert matrix is connected with multiple expert hidden vectors and expert Association relations.Finally,the results of convolution and pooling operations are connected with the hidden vectors of the project,and the matching scores between the experts and the project are obtained by calculating the multi-layer perceptron.(2)Expert recommendation method for multi-projects evaluation that integrates project association and expert AssociationBased on the research of expert recommendation method for single project evaluation,this paper proposes a multi-project evaluation expert recommendation method which combines project association and expert Association on the problem of ranking scores of multi-project recommendation experts by using the relationship between projects and experts.The principle of this method is the same as that of the previous expert recommendation method for single project,but the model constructed by this method is more complex.The project correlation matrix and the expert correlation matrix must be constructed simultaneously,and then connected to the project matrix and the expert matrix respectively by multiple project vectors and expert vectors mapped by Embedding.Finally,the results of the convolution and pooling operations are connected and the matching scores between the experts and the projects are obtained through the multi-layer perceptron.(3)Design and Implementation of Expert Information Management System for Scientific and Technological Talents in Yunnan ProvinceThis paper mainly introduces the design and implementation of the main functions of the expert information management system for scientific and technological talents in Yunnan Province,and focuses on the expert recommendation function.Among them,the implementation of the recommendation model is based on part of open source code published by He et al.and a lot of improvements have been made,including modification of network structure,activation function,loss function and so on.The project association and expert association are integrated into the framework of convolutional neural network.The implementation of expert recommendation function in this system has an important supporting role in the research of expert recommendation method for multi-project evaluation,which integrates project association and expert association.
Keywords/Search Tags:Recommendation System, Expert Recommendation, Probability Diagram Model, Convolutional Neural Network
PDF Full Text Request
Related items