基于改进ADMM算法的图书馆读者借阅数据分类研究  

Research on Borrowing Data Classification of Library Reader Based on Improved ADMM Algorithm

作　　者：周华君[1] Zhou Huajun(Maoming Library,Maoming Guangdong 525000)

机构地区：[1]茂名市图书馆,广东茂名525000

出　　处：《情报探索》2022年第7期92-96,共5页Information Research

摘　　要：[目的/意义]旨在更好地为读者提供个性化的图书推荐服务提供参考。[方法/过程]提出一种基于改进ADMM算法的读者借阅数据分类方法,该方法具有可分离的凸优化特性,同时结合了典型相关分析、多分类逻辑回归的对数似然函数以及Group Lasso损失函数。从而考虑了不同读者及图书间的相似性,并挖掘不同读者数据之间的关联结构。[结果/结论]相较于混合属性算法,该算法在不同数据集下的执行时间提升了5ms~25ms,同时读者借阅分类精度提高了1%~7%,有效提高了图书馆读者数据的利用率。[Purpose/significance]The paper intends to provide references for providing personalized book recommendation service for readers.[Method/process]A reader borrowing data classification method based on improved ADMM(alternative direction method of multipliers)algorithm is proposed,which is combined with canonical correlation analysis,the log likelihood function of multi classification logistic regression,and Group Lasso loss function.Thus,the similarity between different readers and books is considered,and the association structure between different reader data is mined.[Result/conclusion]Compared with the hybrid attribute algorithm,the algorithm in different sets shortens the execution time(5ms~25ms),improves the accuracy(1%~7%)of readers borrowing classification,and increases the utilization of readers data in library effectively.

关 键 词：读者数据 ADMM 关联分析 变量回归 

分 类 号：G258.2[文化科学—图书馆学]