自注意力机制支持下的混合推荐算法

Collaborative filtering is one of the most commonly used methods in recommendation systems. However, factors such as the sparsity of the rating matrix in recommendation systems, the cold start problem, and the fact that most recommendation algorithms often only consider the user's perspective and ignore the relationships between items, limit the practical effectiveness of recommendation algorithms. The paper proposes a deep learning model based on the self-attention mechanism - AS - SADDL, which is used to model user interaction data and learn the representation of users' latent preferences. This model uses a multiple self-attention mechanism to mine the correlation relationships between data from the user's interaction data and learns the representation of users' latent preferences through a deep neural network. At the same time, principal component analysis (PCA) is used to reduce the dimension of the item rating data and calculate the similarity between the item rating data. The similarity between the representation of users' latent preferences and the representation of item characteristics is combined as the final result to recommend items to users. Experiments on real datasets show that the AS - SADDL model has a good computational effect.

协同过滤是推荐系统中最常用的一种方法,但推荐系统中评分矩阵的稀疏性、冷启动性和大多数推荐算法往往只从用户的角度出发忽略了商品间的关系等因素,限制了推荐算法的实际效果.论文提出一种基于自注意力机制(Self-Attention)的深度学习模型--AS-SADDL,用以进行建模用户交互数据及学习用户潜在偏好表示.该模型采用多重自注意力机制从用户的交互数据中挖掘数据间的关联关系,并通过深层神经网络学习用户潜在偏爱表示.同时用主成分分析法(PCA)对项目评分数据进行降维,并计算项目评分数据间的相似性,结合用户潜在偏爱表示与项目特征表示间的相似性作为最终结果,对用户进行项目推荐.在真实数据集上的实验表明,AS-SADDL模型具有较好的计算效果.