基于注意力机制的深度协同推荐模型

Deep Collaborative Recommendation Model Based on Attention Mechanism

针对传统矩阵分解算法无法挖掘深层隐含信息,以及未能充分利用用户和项目评论的问题,提出基于注意力机制的深度协同推荐模型。首先采用注意力机制对评论文本赋权,使用并行的卷积神经网络分别提取用户评论和项目评论特征,同时将评分矩阵输入多层感知机,得到用户隐表示和项目隐表示;然后对两个网络提取的用户特征和项目特征进行融合;最后使用因子分解机和深度神经网络分别提取线性和非线性特征,以进行评分预测。在Amazon的3组公共数据集上进行实验,发现该模型的RMSE达到0.83。与5组对照模型相比,新建模型的RMSE分别降低了14.0%、11.2%、9.8%、7.7%、3.9%,表明该模型能有效提升推荐效果。

In view of the problems that traditional matrix factorization algorithms cannot mine deep hidden information and insufficient utilization of user and item comments,propose a deep collaborative recommendation model based on attention mechanism. Firstly,the attention mechanism is introduced to weight the review text,and the parallel convolutional neural network is used to extract user reviews and item review features respectively,At the same time,the rating matrix is input into the multi-layer perceptron to get the user implicit representation and the item implicit representation. And then integeter the user features and item features. Finally,a factorization machine and a deep neural network are used to extract linear and non-linear features respectively for scoring prediction. The RMSE reached 0.83 on the open data set of Amazon. The RMSE was readuced by 14.0%,11.2%,9.8%,7.7%,3.9% respectively,compared with five groups of models. Experiments showed that the proposed model can improve the recommendation effect effectively.