融合深度神经网络和方面感知的可解释推荐方法

为提升推荐结果的准确性和可解释性，提出一种融合深度神经网络和方面感知的可解释推荐方法。针对评分数据的稀疏性问题，综合考虑显式和隐式评分数据，通过深度神经网络的矩阵分解模型学习用户和物品的潜在特征;通过无监督的方面提取模块来学习用户和物品的方面特征;将潜在特征和方面特征统一到预测层进行评分预测;针对生成解释质量低且缺乏个性化的问题，在评分预测的基础上，采用提取的主题词和预定义的神经模板相结合生成推荐理由，提高解释的生成质量。实验表明，提出的方法不仅能准确预测用户对物品的评分，还能够生成具有解释性的推荐理由，且生成的解释质量优于对比方法。

Fusing deep neural networks and aspect-aware for explainable recommendation method

To improve the accuracy and interpretability of recommendations, a method of interpretability recommendation fusing deep neural network and aspect-aware is proposed. First, for the sparsity problem of rating data, the explicit and implicit rating data are comprehensively considered, and the latent features of users and items are learned through the matrix decomposition of deep neural networks. Second, the aspect features of users and items are learned through the unsupervised aspect extraction module. Then, the latent features and aspect features are integrated into the prediction layer for rating prediction. Finally, for the problems of low explanation quality and lack of personalization, based on the rating prediction, the extracted topic and predefined neural templates are combined to generate recommendation reason to improve the generation quality of explanations. Experimental results on Amazon data sets show that the proposed model can not only accurately predict user ratings of items, but also generate explanatory reasons for recommendations. In comparison, the explanation quality generated by this method is somewhat better than that of the contrastive methods.