融合用户偏好度的图书资源推荐研究  

Research on Book Resource Recommendation Based on User Preference

作　　者：张文德[1,2] 张梦琪 李婵[3] 李海霞 Zhang Wende;Zhang Mengqi;Li Chan;Li Haixia(Institute of Information Management Fuzhou University,Fuzhou Fujian 350108;Fuzhou University Library,Fuzhou Fujian 350108;School of Public Administration&Law Fujian Agriculture and Forestry University,Fuzhou Fujian 350108;Health Bureau of Changle County,Weifang Shandong 262400)

机构地区：[1]福州大学信息管理研究所,福建福州350108 [2]福州大学图书馆,福建福州350108 [3]福建农林大学公共管理与法学院,福建福州350108 [4]昌乐县卫生健康局,山东潍坊262400

出　　处：《情报探索》2023年第5期33-38,共6页Information Research

基　　金：国家自然科学基金“演化博弈视角下视频类知识付费内容著作权侵权风险分析及治理机制研究”(项目编号:71804024);福建省社科研究基地重大项目“‘双碳’目标下福建省乡村自然资源多元共治机制研究”(项目编号:FJ2022MJDZ019);福建省中青年教师教育科研项目(信息化专项)“面向在线教育平台的智能+学习资源精准推荐研究”(项目编号:JAT211018)成果之一。

摘　　要：[目的/意义]传统PersonalRank方法仅考虑读者与图书的描述性特征或单一评分数据,通过结合包含语义信息的交互数据从而更完整体现用户偏好度,并将此融合到传统PersonalRank方法以优化图书资源个性化推荐的效果。[方法/过程]利用BERT模型评估读者评论数据中的情感分值,并融合读者原始评分数据以度量用户偏好度;利用包含用户偏好度的读者交互数据构建读者—图书二部图;基于构建的二部图,利用PersonalRank方法捕捉读者与图书的潜在相关性,从而进行图书资源推荐。使用豆瓣图书的评分评论数据集,将本文方法与原PersonalRank方法和ItemCF方法进行对比。[结果/结论]本文所用方法相较于对比方法在准确率和召回率上均有提升,证明该方法能够提高图书资源个性化的效果。[Purpose/significance]The traditional PersonalRank method only considers the descriptive characteristics or single rating data of readers and books.This paper combines interaction data containing semantic information to more completely represent user preferences and integrates this into the traditional PersonalRank approach to optimize the effectiveness of personalized recommendation of book resources.[Method/process]The BERT model is used to evaluate sentiment scores in readers’comment data and then the readers’original rating data is integrated to measure user preferences.The reader interaction data containing user preferences is used to construct a reader-book bipartite graph.Based on the constructed bipartite graph the potential relevance of readers and books is captured using the PersonalRank method for book resource recommendation.Using the rating and comment dataset of Douban books the method in this paper is compared with the original PersonalRank method and ItemCF method.[Result/conclusion]The method used in this paper improves the accuracy and recall rate compared with the comparison method which proves that this method can improve the personalized effect of book resources.

关 键 词：随机游走 PersonalRank 二部图 用户偏好度 图书资源 

分 类 号：G250[文化科学—图书馆学]