摘要
针对推荐系统的自然噪声问题,提出一种基于栈式降噪自编码器的深度推荐算法(Deep Recommendation Algorithm Based on Stacked Denoising Auto-encoder,SDAE-DR)。该推荐算法包含采样、重构及推荐3个模块。采样模块利用交互矩阵中的隐式反馈和项目的属性信息构建知识图谱,利用知识图谱中项目之间共同的知识实体来执行负采样。重构模块利用采样模块初步筛选的数据获得用户和项目评分向量,利用栈式自编码器对数据进行重构,从而获得用户和项目的隐表示。推荐模块利用用户信息和项目信息分别获取用户和项目特征向量,再与重构模块获得的隐表示结合,通过多层感知机来获得预测评分。实验表明,该算法与基准线相比,具有更高的推荐准确性和算法运行效率。
Aiming at the problem of natural noise in the recommendation system,a Deep Recommendation Algorithm Based on Stacked Denoising Auto-encoder(SDAE-DR)is proposed.The recommendation algorithm includes three modules:sampling,reconstruction and recommendation.The sampling module uses the implicit feedback in the interaction matrix and the attribute information of the items to construct the knowledge graph,and uses the knowledge entities common among the items in the knowledge graph to perform negative sampling.The reconstruction module uses the data preliminarily screened by the sampling module to obtain user and item rating vectors,and uses the stacked autoencoder to reconstruct the data to obtain the implicit representation of users and items.The recommendation module uses the user information and item information to obtain the user and item feature vectors respectively,and then combines with the implicit representation obtained by the reconstruction module to obtain the prediction score through the multi-layer perceptron.Experiments show that the algorithm has higher recommendation accuracy and algorithm operation efficiency compared with the baseline.
作者
钟裔灵
朵琳
ZHONG Yiling;DUO lin(Faculty of Information Engineering and Automation,Kunming University of Science and Technology,Kunming 650500,China)
出处
《电视技术》
2022年第7期60-64,共5页
Video Engineering
基金
国家自然科学基金项目(No.61761025
No.61962032)。
关键词
自然噪声
知识图谱
自编码器
深度学习
推荐系统
natural noise
knowledge graph
autoencoder
deep learning
recommendation system
