基于深度学习的矩阵分解推荐模型

Matrix factorization recommender system based on deep learning

通过对深度学习和矩阵分解技术进行结合,设计一个深度神经网络对用户和物品进行特征提取,形成用户隐向量和物品隐向量的方法,计算这两个隐向量的内积得到用户对物品的评分预测。为提高推荐精度,提出使用显式数据和隐式数据并设计新的损失函数能够同时计算这两类数据损失的方法。在两个公开数据集上的实验结果表明,该方法比基线模型在HR和NDCG推荐指标上有较大提升。

A deep neural network was designed by the combination of deep learning and matrix factorization to extract the features of users and items to form user latent vector and item latent vector,and the rating prediction was obtained by calculating the inner product of these two latent vectors.To improve the accuracy of recommendation,explicit data and implicit data were used in the course of designing a new loss function to calculate the loss of data of these two types.Results of experiments on two different datasets show that a greater improvement in the evaluation of HR and NDCG indicators compared to the baseline model is demonstrated.