一种基于深度学习的混合推荐算法  

Hybrid Recommendation Algorithm Based on Deep Learning

作　　者：曾旭禹 杨燕[1,2] 王淑营 何太军[1] 陈剑波 ZENG Xu-yu;YANG Yan;WANG Shu-ying;HE Tai-jun;CHEN Jian-bo(School of Information Science and Technology,Southwest Jiaotong University,Chengdu 611756,China;Key Lab of Cloud Computing and Intelligent Technology,Sichuan Province,Chengdu 611756,China)

机构地区：[1]西南交通大学信息科学与技术学院,成都611756 [2]四川省云计算与智能技术高校重点实验室,成都611756

出　　处：《计算机科学》2019年第1期126-130,共5页Computer Science

基　　金：国家自然科学基金(61572407);国家科技支撑计划(2015BAH19F02)资助

摘　　要：推荐系统在电子商务的发展中发挥着越来越重要的作用,但用户对物品评分数据的稀疏性往往是推荐精度较低的重要原因。目前通常采用推荐技术对辅助信息进行处理,以缓解用户评价的稀疏性,并提高预测评分精度。通过相关模型,可以利用文本数据来提取物品的隐藏特征。最近,深度学习算法快速发展,因此文中选用了一种具有强大特征提取能力的新型深度网络架构——变分自编码器(Variational AutoEncoder,VAE)。通过将无监督变分自编码融合到概率矩阵分解(Probability Matrix Factorization,PMF)中,构建了一种感知上下文的新型推荐模型——变分矩阵分解(Variational AutoEncoder Matrix Factorization,VAEMF)。首先使用TD-IDF对物品的评价文档进行数据预处理,然后对处理后的数据使用VAE捕获物品的上下文信息特征,最后使用概率矩阵分解进一步提高预测评分精度。在两个真实数据集上的实验结果验证了所提方法相较于自编码算法及概率矩阵分解算法的优势。Recommendation system is playing an increasingly indispensable role in the development of e-commerce,but the sparsity of user’s rating data for the items in the recommendation system is often an important reason for the low recommendation accuracy.At present,the recommendation technology is usually used to process the auxiliary information to alleviate the sparsity of the user evaluation and improve the accuracy of the prediction score.Text data can be used to extract the hidden features of the item through related models.In recent years,the deep learning algorithm has developed rapidly.Therefore,this paper chose a variational autoencoder(VAE),which is a new type of network structure with powerful feature extraction capabilities.This paper proposed a novel context-aware recommendation model integrating the unsupervised method VAE into the variable matrix factorization(VAEMF)in the probability matrix factorization(PMF).Firstly,TD-IDF is used to preprocess the evaluation documents of the item.Then,the VAE is utilized to capture the context information features of the item.Finally,the probability matrix factorization is used to improve the accuracy of the prediction score.The experimental results on two real data sets show that this method is superior to the autoencoder and the probability matrix factorization recommendation methods.

关 键 词：推荐系统 深度学习 变分自编码 矩阵分析 

分 类 号：TP391.3[自动化与计算机技术—计算机应用技术]