基于多核学习卷积神经网络的稀疏数据推荐  

Sparse data recommendation based on multiple kernel learning convolutional neural networks

作　　者：霍雨佳 左欣 张虹 HUO Yu-jia;ZUO Xin;ZHANG Hong(School of Artificial Intelligence and Electrical Engineering,Guizhou Institute of Technology,Guiyang 550003,China;Special Key Laboratory of Artificial Intelligence and Intelligent Control,Provincial Department of Education,Guiyang 550003,China;Continuing Education College,Guizhou Education University,Guiyang 550018,China;School of Foreign Languages,Guizhou University of Finance and Economics,Guiyang 550025,China)

机构地区：[1]贵州理工学院人工智能与电气工程学院,贵州贵阳550003 [2]贵州省教育厅贵州省人工智能与智能控制特色重点实验室,贵州贵阳550003 [3]贵州师范学院继续教育学院,贵州贵阳550018 [4]贵州财经大学外语学院,贵州贵阳550025

出　　处：《计算机工程与设计》2021年第2期489-496,共8页Computer Engineering and Design

基　　金：现代制造技术教育部重点实验室开放课题基金项目(黔教合KY字[2018]477号);贵州省高等学校教学内容和课程体系改革基金项目(2017520072);贵州省大数据统计分析重点实验室基金项目(黔科合平台人才[2019]5103号)。

摘　　要：针对当前稀疏数据推荐准确率低的问题,提出一种基于多核学习卷积神经网络的稀疏数据推荐算法。将项目的辅助信息送入卷积神经网络学习特征,将向量在可再生核希尔伯特空间组合,利用多核学习技术增强卷积神经网络的特征学习能力;基于学习的卷积特征集初始化非负矩阵模型,通过非负矩阵模型实现对缺失评分的预测。实验结果表明,该算法有效提高了稀疏数据集的推荐性能,验证了多核学习卷积神经网络的有效性。Aiming at the problem of low accuracy in existing sparse data recommendation systems,a sparse data recommendation system based on multiple kernel convolutional neural networks was proposed.The auxiliary information of items was fed into convolutional neural networks to learning their features,vectors were combined in reproducing kernel Hilbert space,and multiple kernel learning technique was adopted to enhance the feature learning ability of convolutional neural networks.The learned convolutional features were used to initialize the non-negative matrix model,the losing ratings were predicted using the non-negative matrix model.Experimental results show that the proposed algorithm improves the recommendation performance for sparse data effectively,meanwhile the proposed multiple kernel learning convolutional neural networks is validated.

关 键 词：稀疏数据 推荐系统 评分预测 卷积神经网络 多核学习 项目上下文 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]