基于PU-Learning和TextCNN的文献推荐方法研究
Research on Literature Recommendation Methods Based on TextCNN and PU-Learning
刁羽 1薛红1
作者信息
- 1. 四川轻化工大学图书馆 四川自贡,643000
- 折叠
摘要
论文旨在将现有的机器学习研究成果运用到图书馆文献推荐的实际工作中,以充分发挥电子资源的作用.鉴于难以获得用户对文献资源的显式评价,因此将用户浏览、下载的文献视为正类文献,将用户未交互的文献视为未标记文献,通过卷积网络文本分类模型并结合PU-Learning算法对待推荐文献的推荐概率进行预测.实践证明该方法具有较高的精准性,能够在图书馆文献推荐实际应用中发挥作用.
Abstract
This paper aims to apply existing machine learning research to the practical work of library literature recommendation,in order to make full use of electronic re-sources.Due to the difficulty in obtaining users'explicit ratings on literature re-sources,the literature browsed and downloaded by users is treated as the positive,and literature without user interaction is treated as unlabeled.The recommendation probability of candidate literature is predicted through TextCNN classification model combined with PU-Learning algorithm.Practice has proved that this method has high accuracy and can play a role in the actual application of library literature recommendation.
关键词
卷积神网络/电子文献推荐/PU-Learning/文本分类Key words
Convolutional neural network/Electronic literature recommendation/PU-Learning/Text classification引用本文复制引用
基金项目
四川省高等学校人文社会科学重点研究基地——新建院校改革与发展研究中心项目(XJYX2019B05)
出版年
2024
NETL
NSTL
维普
万方数据