In heterogeneous e-commerce recommender systems, the type and attribute information of users and products contain rich semantics, which can benefit the prediction and explanation of user ratings of interesting items. Existing studies include collaborative and content-based recommendations that mainly capture semantic features by considering user–item interactions or behavioral history records, which ignores the explanatory role of the product type and attribute. In this paper, we first propose an attentional attribute and interaction method used to model the semantic embeddings of users and items. We then construct a type-specific matrix to exploit heterogeneous type-specific information to learn user and item representations. The incorporated heterogeneous type information helps capture a user’s latent features that solve the sparsity problem of user–item interactions for the recommender systems. Further, the rating relationship of the nodes is predicted through the translation mechanism based on user and items’ type-specific representations. Extensive experimental results on real-world datasets demonstrate the superior performance of the proposed model over several state-of-the-art methods and show the visual interpretability for rating behaviors in e-commerce recommender systems.

在异构电子商务推荐系统中，用户和产品的类型和属性信息包含丰富的语义，这有助于对有趣项目的用户评分进行预测和解释。现有研究包括基于协作和基于内容的建议，这些建议主要通过考虑用户与项目的交互或行为历史记录来捕获语义特征，而忽略了产品类型和属性的解释作用。在本文中，我们首先提出一种注意力属性和交互方法，用于对用户和项目的语义嵌入进行建模。然后，我们构造一个特定于类型的矩阵，以利用异构的特定于类型的信息来学习用户和项目表示。合并的异构类型信息有助于捕获用户的潜在功能，从而解决推荐系统中用户与项目交互的稀疏性问题。此外，通过基于用户和项的类型特定表示的转换机制来预测节点的评级关系。真实数据集上的大量实验结果证明了所提出的模型优于几种最新方法的性能，并显示了电子商务推荐系统中评分行为的可视解释性。