摘要
【目的/意义】利用距离分解改进矩阵分解方法的限制,优化图书个性化推荐的效果。【方法/过程】将传统用户-项目评分矩阵转换为用户-项目距离矩阵;利用距离分解方法获得用户-项目间的距离关系以替代原有方法中的相似度关系;利用深度学习框架实现该方法并进行图书个性化推荐。【结果/结论】利用豆瓣图书评分数据作为数据源分别对本文所用方法以及对照方法进行实验对比。结果表明,本文所用方法相较于对照方法在RMSE和MAE上均有提升,从而证明该方法能够提高图书个性化推荐的效果。【创新/局限】将深度距离分解方法应用到图书个性化推荐,从而优化了图书个性化推荐的效果。但方法仅在一个真实数据集上进行实验,在接下来还需要在更多数据集上进行验证。
【Purpose/significance】This study tries to improve the limitation of matrix factorization and optimize the effect of book personalized recommendation based on deep metric factorization.【Method/process】First, we convert the rating matrix into distance metric matrix. Then, calculating the distance between user and item based on Metric Factorization method. Finally, we use deep learning framework to build this model.【Result/conclusion】We use Douban book rating dataset to train and test this model, meanwhile we compare our approach with three baselines. The results show that our approach is better than other baselines on RMSE and MAE. And the experiment results show that this approach can improve the effect of book personalized recommendation.【Innovation/limitation】The effect of book personalized recommendation has been optimized based on deep metric factorization. But this method is only experimented on a real dataset, and it needs to be tested on more dataset in the future.
作者
黄禹
张文德
张诗雨
HUANG Yu;ZHANG Wen-de;ZHANG Shi-yu(Institute of Information Management,Fuzhou University,Fuzhou 350108,China;Library of Fuzhou University,Fuzhou 350108,China)
出处
《情报科学》
CSSCI
北大核心
2021年第3期76-81,共6页
Information Science
基金
2017年赛尔网络下一代互联网创新项目“融合评论标签的个性化学习资源推荐关键技术研究”(NGII20170522)
2018年赛尔网络下一代互联网创新项目“面向慕课课程评论的情感标签信息抽取关键技术研究”(NGII20180509)。
关键词
深度学习
距离分解
矩阵分解
图书个性化推荐
deep learning
metric factorization
matrix factorization
book personalized recommendation