摘要
变分自编码器是一种非常简洁有效的非监督学习方法,应用在推荐系统领域也能取得极佳的性能。推荐系统的主要工作之一是对缺失的数据进行估计并补全,变分自编码器通过对已有数据的学习和抽象能够挖掘出数据间隐式的关联因子,并基于此完成对缺失数据的预测。该文将额外的辅助信息加入到变分自编码器中以提高预测的准确度,并通过在包括高考成绩及电影评分等在内的实际数据集测试中验证了辅助信息的有效性,当辅助信息充足时在高考成绩数据集上最多可以降低31%的均方根误差。
The variational autoencoder（VAE）unsupervised learning method can provide excellent results in recommendation systems.Recommendation systems seek to accurately identify a missing value with the VAE learning a latent factor from the input and then predicting when to use this for reconstructing the result.Side information was added to the VAE to improve the predictions with tests on datasets including MovieLens and grades data showing that it can significantly improve the prediction accuracy by up to 31% with enough side information with the grades dataset.
作者
刘卫东
刘亚宁
LIU Weidong;LIU Yaning(Department of Computer Science and Technology, Tsinghna University, Beijing 100084, China)
出处
《清华大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2018年第8期698-702,共5页
Journal of Tsinghua University(Science and Technology)
关键词
推荐系统
变分推理
自编码器
协同过滤
recommendation systems
variational inference
autoencoder
collaborative filtering
