面向新读者和新图书的数字图书馆个性推荐冷启动问题研究  

Research on Cold-start Problem of Personal Recommendation in Digital Library Oriented to New Readers and New Books

作　　者：凌霄娥[1] 周兵[2] 李克潮[3] 

机构地区：[1]广西民族师范学院图书馆,广西崇左532200 [2]广西卫生职业技术学院图书馆,广西南宁530023 [3]广西中医药大学附属瑞康医院计算机中心,广西南宁530011

出　　处：《情报理论与实践》2014年第8期100-104,99,共6页Information Studies:Theory & Application

基　　金：广西教育厅科研项目"个性化推荐冷启动及稀疏性关键技术研究"(项目编号:201204LX481);CALIS广西壮族自治区文献信息服务中心预研项目"云计算环境下高校读者模糊借阅偏好的图书推荐研究"(项目编号:CALISGX201305)的研究成果

摘　　要：针对高校图书馆新读者、新图书因没有借阅历史记录导致的冷启动问题,借鉴电子商务推荐系统冷启动问题的一些解决办法,提出利用改进的K-medoids分别对已有读者、已有图书进行基于借还时间间隔的聚类,根据已有读者和已有图书的特征分别建立决策树分类模型。然后,通过目标新读者、新图书的特征分别匹配对应的已有读者类、已有图书类。依据所归属的类内借阅偏好情况,向目标新读者推荐其可能喜好并且在架的图书,并把目标新图书推荐给相应的读者。在高校图书馆借阅记录数据集上,提出的方法得到有效的验证。Based on the problem of cold-start that led by lack of new readers and new books borrow history in university library, this paper draws lessons from some cold-start problem solutions of e-commerce recommendation system. The paper proposes an improved K-medoids algorithm to cluster existing readers and books respectively based on borrow-return time interval. Moreover, the paper establishes decision tree classification model for the existing readers and books respectively based on their features. Then, the paper matches the corresponding categories of existing readers and books by their features. Based on the borrow preferences in the class belong, the paper recommends books in the frame to the target new readers that they may prefer, and recommends target new books to the appropriate readers. The proposed method was effectively verified based on the borrow records dataset of university library.

关 键 词：新读者 新图书 图书推荐 冷启动 决策树 

分 类 号：G250.76[文化科学—图书馆学]