摘要
为解决单一推荐算法应用具有局限性和用户行为数据具有稀疏性的问题,将迁移学习方法应用到组合推荐算法.该算法首先分别利用矩阵分解(MF)推荐算法和深度神经网络(DNN)推荐算法对用户行为数据进行预测,然后利用迁移学习方法将训练出来的特征数据作为组合推荐算法的输入,并进行再次训练,获得预测评分,实现对目标用户的推荐.实验结果表明,具有迁移学习的基于矩阵分解和深度神经网络的组合推荐算法能够有效地提升推荐质量.
In order to solve the limitation of single recommendation algorithm application and the sparsity of user behavior data, this paper applies the transfer learning method to the combined recommendation algorithm. Firstly, the paper respectively utilizes the matrix decomposition (MF) recommendation algorithm and deep neural network (DNN) recommendation algorithm to predict user behavior data. Then the paper uses the trained feature data as the input of the combined recommendation algorithm via the transfer learning method, and next carries out the training again to obtain the forecast score and realize recommendations to target users. The experimental results show that the combined recommendation algorithm based on matrix decomposition and deep neural network with transfer learning can effectively improve the quality of recommendation.
作者
申远
黄志良
蒋苏蓉
胡彪
SHEN Yuan;HUANG Zhiliang;JIANG Surong;HU Biao(Air Force EarlyWarning Academy,Wuhan 430019, China)
出处
《空军预警学院学报》
2019年第2期136-141,共6页
Journal of Air Force Early Warning Academy
关键词
组合推荐
迁移学习
矩阵分解
深度神经网络
combined recommendation
transfer learning
matrix decomposition (MF)
deep neural network (DNN)
