基于NCF框架的CNMF模型在推荐领域的研究  

Research on CNMF Model Based on NCF Framework in the Field of Recommendation

作　　者：崔一航 

机构地区：[1]天津工业大学计算机科学与技术学院,天津

出　　处：《计算机科学与应用》2023年第10期1902-1910,共9页Computer Science and Application

摘　　要：深度学习在推荐系统领域取得了显著的成功,然而,现有的一些模型仍存在一些局限性,如不能捕捉用户和物品之间的非线性交互信息。为了解决这个问题,我们对NeuMF模型进行了改进,提出CNMF模型。首先,我们将NeuMF模型中的多层感知机(MLP)层替换为卷积神经网络(CNN)。CNN能够有效地提取用户和物品之间的时空特征,从而更好地捕捉它们之间的关系。其次,我们引入了注意力机制来进一步增强模型的性能。注意力机制可以自动学习用户和物品之间的重要关系,从而更好地建模推荐过程。我们通过计算用户和物品之间的注意力权重来加权池化其交互特征,从而更准确地预测用户的喜好。最后,我们在经典的推荐数据集上进行了大量的实验。实验结果表明,我们提出的改进算法在准确性和效率方面显著优于传统的NeuMF模型。特别是在用户和物品数量较大的情况下,我们的算法展现出更好的稳定性和可扩展性。Deep learning has achieved remarkable success in the field of recommendation systems, however, some existing models still have some limitations, such as the inability to capture non-linear interaction information between users and items. To solve this problem, we improved NeuMF model and proposed CNMF model. First, we replace the multi-layer perceptron (MLP) layer in the NeuMF model with a convolutional neural network (CNN). CNN can effectively extract the spatio-temporal characteristics between users and items, so as to better capture the relationship between them. Secondly, we introduce the attention mechanism to further enhance the performance of the model. Attention mechanisms can automatically learn important relationships between users and items to better model the recommendation process. We can more accurately predict user preferences by calculat-ing the weight of attention between users and items to pool their interaction characteristics. Finally, we conducted a large number of experiments on classical recommendation datasets. Experimental results show that our improved algorithm is significantly superior to the traditional NeuMF model in terms of accuracy and efficiency. Especially in the case of a large number of users and items, our algorithm shows better stability and scalability.

关 键 词：深度学习 推荐算法 神经网络 特征 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]