融合注意力机制和评论文本特征的推荐算法

Recommendation algorithm combining attention mechanism and review feature

为区分基于评论文本推荐算法中不同评论文本数据对不同用户或物品的差异,提出一种融合注意力机制和评论文本特征的推荐算法(RAAM)。在卷积神经网络中加入三级注意力机制,分别从单词级别、语句级别和评论级别为不同用户或物品区分评论数据的重要性,引入共同注意力网络模拟用户与物品之间的交互,获取更多用户和物品的交互信息,提高推荐效果。在Amazon的5个数据集上的对比实验结果验证了算法的有效性。

To distinguish the importance of different review data for different users and items in the recommendation algorithm based on review,a recommendation algorithm combining attention mechanism and review feature was proposed(RAAM).A three-level attention mechanism was added into the convolutional neural network to distinguish the importance of comments for users or items from word level,sentence level and comment level.A co-attention network was introduced to simulate the interaction between the user and the item,so more fine-grained interaction characteristics were obtained.Experimental results on five Amazon data sets verify the effectiveness of the recommendation algorithm.