Computer Engineering and Applications ›› 2019, Vol. 55 ›› Issue (19): 178-184.DOI: 10.3778/j.issn.1002-8331.1811-0218

Previous Articles     Next Articles

Improved Model Combining Improved Matrix Decomposition and Convolutional Neural Networks

CAI Nian, LIU Guangcong, CAI Hongdan   

  1. 1.School of Computer, Guangdong University of Technolog, Guangdong 510006, China
    2.School of Computer, Sanxia University, Yichang, Hubei 443002, China
  • Online:2019-10-01 Published:2019-09-30



  1. 1.广东工业大学 计算机学院,广州 510006
    2.三峡大学 计算机学院,湖北 宜昌 443002

Abstract: Model research and improvement are carried out on issues such as sparse recommendation data scoring and evaluation information explosion. Firstly, based on the traditional matrix decomposition model, the paper adds the influence factors of users and projects to improve the generalization ability of the prediction model. Secondly, this paper establishes a cross-channel convolutional neural network to identify user evaluation information, combines the improved matrix decomposition model with the improved convolutional neural network, and proposes a combination of improved matrix decomposition and cross-channel convolutional neural network, model to improve the accuracy of the predictive model. The experimental results show that the optimal performance of the model is 2.96%, 10.27% and 1.77% higher than that of PMF, CTR and CDL in three data sets respectively. The performance of MF&CNN is increased by 0.29%, 2.98% and 0.08% respectively. When the data density increases from 20% to 80%, the model predicts performance will be further improved.

Key words: recommended system, Information explosion, matrix factorization, convolution neural network

摘要: 针对推荐系统评分数据稀疏和评价信息爆增等问题进行模型研究和改进。在传统矩阵分解模型基础上加入了用户和项目的影响因子,提高预测模型的泛化能力;建立跨通道卷积神经网络对用户评价信息进行识别,将改进的矩阵分解模型与改进卷积神经网络进行结合,提出一种改进矩阵分解与跨通道卷积神经网络结合的推荐模型,提高预测模型的准确度。实验结果表明,该模型预测性能相对于PMF、CTR和CDL在三个数据集上的最优性能分别提升2.96%、10.27%和1.77%,相对于MF&CNN性能分别提升0.29%、2.98%和0.08%;当数据密度从20%增至80%时,模型预测性能会进一步提升。

关键词: 推荐系统, 信息爆增, 矩阵分解, 卷积神经网络