基于用户画像的高校采购评审专家推荐算法

Expert recommendation algorithm for university procurement review based on user portraits

该文设计了一种可调参的混合推荐算法,融合AUC-MF算法与基于专家画像及项目画像的抽取算法,以提高高校随机抽取评审专家的专业匹配度,解决数据稀疏和冷启动问题。该文以某高校近6年政府采购项目历史评审数据以及项目文本信息构建数据集,对算法进行验证,该算法可有效缓解高校自选评审专家过程中的数据稀疏和冷启动问题,提高随机抽取专家的专业匹配度。

[Objective]With the advancement of“government function reform”in the scientific research field,central universities and research institutes can now select experts for government procurement projects involving complex technologies.Currently,two expert selection methods are mainly used:self-selection and database recommendation.The self-selection method enhances the autonomy of university procurement but also brings certain integrity risks.The database recommendation method relies on the precision of the professional settings in the review expert database.It may face issues such as reduced recommendation efficiency or mismatches in expertise.However,it effectively reduces the risk of corruption.[Methods]Considering the correlation and continuity of expertise among self-selected review experts in universities,this study aims to enhance the matching accuracy of the database recommendation method and address data sparsity issues and the cold start problem.A tunable hybrid recommendation algorithm was designed by combining the area under the curve AUC-MF model with an expert and project portrait-based recommendation method.The AUC-MF algorithm converts the recommendation problem into a classification problem by treating reviewed and unreviewed projects as positive and negative samples.Furthermore,it converts the data sparsity problem into a data imbalance problem,improves the matrix factorization algorithm by maximizing the AUC,and optimizes the gradient descent algorithm using LambdaMF.It can better solve the data sparsity problem in the expert sample of the university procurement project review.The recommendation algorithm models expert and project portraits using text mining,incorporating dynamic attributes(project reviews)and static attributes(expertise)for expert portraiture and using project names and summaries for project portraiture.Combining expert and project portraits as input and applying logistic regression for recommendations substantially addresses the cold start problem.[Results]Experimental results using historical review data from government procurement projects over the past six years and project text information from a certain university demonstrate the following features.①At K=3,5,and 10(where K represents the top K recommended experts),the F1 scores of the recommendations are 0.5144,0.5712,and 0.4545,respectively.②The dimensionality of portrait vectors impacts recommendation outcomes with optimal dimensions identified between 50 and 60 for K=3,5,and 10.③The parameterα(whereαrepresents the weight of the recommendation algorithm)also impacts recommendation effectiveness,with the best result atα=0.7.④The effectiveness of the proposed algorithm is validated on five practical university projects.[Conclusions]To address the challenge of low professional matching in the recommendation of review experts for university procurement projects,this paper leverages historical project review data and textual information to design an algorithm based on the AUC-MF model,expert portraits,and project portraits.This integrated approach effectively reduces issues related to data sparsity and cold starts,enhances the expert recommendation process,and improves the user experience for expert recommendation.