基于注意力机制的深度协同推荐模型构建

Deep collaborative recommendation model based on attention mechanism

协同过滤中的矩阵分解算法在推荐系统中应用广泛,但其很难充分利用用户和项目描述文档信息,存在显式数据稀疏的问题。由于注意力机制与深度神经网络强大的特征学习能力,本文提出了基于注意力机制的深度协同推荐模型,将评分矩阵的行和列融合用户和项目文档中的描述信息后作为输入,以此缓解数据稀疏性问题;引入注意力机制,有利于神经网络层更好地提取输入向量的非线性特征。在公开的数据集上的实验表明,本文所提出的算法提高了评分预测的准确性。

The matrix decomposition algorithm in collaborative filtering is widely used in the recommendation system,but it is difficult to make full use of the information in the user and item description documents,causing the problem of explicit data sparseness.Due to the powerful feature learning ability of the attention mechanism and deep neural network,this paper proposes a deep collaborative recommendation model based on the attention mechanism,which combines the rows and columns of the scoring matrix with the description information in the user and item documents as input to ease the problem of data sparsity.The attention mechanism is helpful for the neural network layer to better extract the nonlinear characteristics of input vectors.Experiments on the public dataset show that the algorithm proposed in this paper improves the accuracy of scoring prediction.