图书情报工作 ›› 2018, Vol. 62 ›› Issue (3): 104-113.DOI: 10.13266/j.issn.0252-3116.2018.03.013

• 知识组织 • 上一篇    下一篇

基于LDA主题模型的标签混合推荐研究

熊回香, 窦燕   

  1. 华中师范大学信息管理学院 武汉 430079
  • 收稿日期:2017-08-25 修回日期:2017-11-19 出版日期:2018-02-05 发布日期:2018-02-05
  • 作者简介:熊回香(ORCID:0000-0001-9956-3396),教授,博士,E-mail:hxxiong@mail.ccnu.edu.cn;窦燕(ORCID:0000-0003-0029-2624),硕士研究生。
  • 基金资助:
    本文系国家社会科学基金项目"大众分类中标签间语义关系挖掘研究"(项目编号:12BTQ038)研究成果之一。

Research on Tag Hybrid Recommendation Based on LDA Topic Model

Xiong Huixiang, Dou Yan   

  1. School of Information Management, Central China Normal University, Wuhan 430079
  • Received:2017-08-25 Revised:2017-11-19 Online:2018-02-05 Published:2018-02-05

摘要: [目的/意义]针对目前使用标签推荐方法所得结果不理想的问题,改进传统相似度计算方式,并结合多种标签推荐方法,提高推荐准确性。[方法/过程]融合基于内容与协同过滤的推荐思想,利用LDA进行相似度计算得出资源与用户的近邻集合,并抽取资源内容关键词,以此构建标签混合推荐模型,最后以"豆瓣读书"为例对模型进行验证,同时与几种标签推荐方法进行比较。[结果/结论]在社会标注系统中,必须考虑用户-资源-标签3个维度,仅考虑单一角度势必会造成结果的不完整,同时在相似度计算时引入LDA能够挖掘潜在语义关系,提高推荐质量,且组合多种方法取长补短可以令推荐结果更为满意。

关键词: 社会标注, 标签推荐, 协同过滤, LDA

Abstract: [Purpose/significance]For the current tag recommendation methods' results not satisfied, this paper aims to improve the traditional similarity calculation method and combine a variety of tag recommendation methods to improve the recommended accuracy.[Method/process]Based on the idea of content and collaborative filtering, LDA is used to calculate the similarity then find the neighbor of resources and users, and combine keywords which are extracted from resource contents to construct the tag hybrid recommendation model. Finally, "Douban reading" is taken as an example to verify the model's effectiveness and compared with several tag recommendation methods.[Result/conclusion]In the social tagging system,three dimensions including user, resource and tag should be considered.Only from one single angle will inevitably cause incomplete results.At the same time, the introduction of LDA in similarity calculation can exploit the potential semantic relation and improve the recommended quality. And the combination of a variety of ways to learn from each other can make the results more satisfactory.

Key words: social tagging, tag recommendation, collaborative filtering, LDA

中图分类号: