融合深度去噪自编码器和注意力机制的推荐算法  

RECOMMENDATION ALGORITHM COMBINING DEEP DENOISING AUTOENCODERAND ATTENTION MECHANISM

作　　者：张卫国[1] 袁炜轩 周熙然 Zhang Weiguo;Yuan Weixuan;Zhou Xiran(Collge of Computer Science and Technology,Xi’an University of Science and Technology,Xi’an 710000,Shaanxi,China)

机构地区：[1]西安科技大学计算机科学与技术学院,陕西西安710000

出　　处：《计算机应用与软件》2023年第8期283-290,共8页Computer Applications and Software

基　　金：国家自然科学基金项目(61902311)。

摘　　要：传统推荐算法无论在特征提取还是相似度计算方面仍存在数据稀疏和大量噪声数据问题,导致推荐效率不高、用户满意度低等问题,由此提出一种融合深度去噪自编码器和注意力机制的推荐算法。将深度去噪自编码器融入到基于项目相似度的协同过滤推荐算法中,同时加入了注意力机制,以惩罚活跃用户对实验结果的影响,既可以挖掘到用户与项目的线性特征又可以学习到用户与项目非线性特征。实验选取了MovieLens和Pinterest两个公开数据集,与传统推荐算法和近些年较先进算法相比,该算法能够显著提升传统推荐算法的性能,并可以缓解传统推荐算法存在的数据稀疏和冷启动问题。Traditional recommendation algorithms still have data sparseness and a large amount of noisy data in both feature extraction and similarity calculation,resulting in low recommendation efficiency and low user satisfaction.Therefore,a recommendation algorithm combining deep denoising autoencoder and attention mechanism is proposed.The deep denoising autoencoder was integrated into the collaborative filtering recommendation algorithm based on item similarity,and the attention mechanism was added to punish the influence of active users on the experimental results,which could not only mine the linear characteristics of users and items,but also learn the non-linear characteristics of them.Two public data sets MovieLens and Pinterest were selected in the experiment.The results show that compared with traditional recommendation algorithms and more advanced algorithms in recent years,this algorithm can significantly improve the performance of traditional recommendation algorithms,and can alleviate the data sparseness and cold start problem of traditional recommendation algorithms.

关 键 词：推荐算法 去噪自编码器 注意力机制 协同过滤 数据稀疏 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]