基于用户评论的动态方面注意力电商推荐深度学习模型

为了提供个性化推荐,推荐系统会将用户和物品分别表达为用户偏好向量和物品特征向量。物品特征向量中不同维度分别对应物品不同的特征。用户偏好向量中各维度表示用户对物品对应维度(特征)的喜好程度。目前大部分的推荐算法都假设为对于不同物品、同一用户的偏好向量是相同的。然而在现实生活中,该假设是不成立的。为此,提出一种结合注意力机制的深度学习模型,其能根据不同的用户-物品对,相应地学习到一个注意力权重向量,最终达到动态调整用户偏好向量的目的。在3组公开数据集上进行对比实验,以预测评分的均方误差(MSE)作为评估指标,实验结果表明该方法比已有的相关算法的效果更好。

A DEEP LEARNING MODEL OF DYNAMIC ASPECT ATTENTION E-COMMERCE RECOMMENDATION BASED ON USER COMMENTS

In order to provide personalized recommendations,the recommended system will express users and items as user preference vector and item feature vector respectively.Different dimensions in the item feature vector respectively correspond to different features of the item,and each dimension in the user preference vector represents the user s preference for the corresponding dimension(feature)of the item.Most of the current recommendation algorithms assume that the user s preference vector is the same for different items.However,this assumption is not true in real life.Therefore,this paper proposes a deep learning model combined with attention mechanism,which can learn an attention weight vector according to different user-item pairs,and finally achieve the purpose of dynamically adjusting the user preference vector.A comparative experiment was carried out on three groups of open datasets,and the MSE of prediction score was taken as the evaluation index.The experimental results show that the method is better than the existing related algorithm.