基于动态二分网络表示学习的推荐方法

构建用户—项目交互网络并学习其表征是一种有效的推荐方法。已有的方法大多将交互网络视为静态同质网络，忽略了交互时序性和节点异质性的影响。针对这一问题，提出一种基于动态二分网络表示学习的推荐方法，首先构建时序加权二分网络；然后将用户节点和项目节点分别映射到不同的向量空间以保留网络的异质性，选择图卷积网络来聚合节点的一阶和高阶邻居信息；最后使用多层感知机学习两类节点嵌入的非线性关系并进行top-N推荐。在Amazon和Taobao数据集上的实验结果表明，该方法在HR和NDCG推荐指标上均显著优于相关的基于静态、异质网络表示学习的方法。

Recommendation method based on dynamic bipartite network representation learning

Representation learning of a user-item interaction network becomes an effective recommendation method. Most of the existing methods regard the interaction network as a static homogeneous network, ignoring the impact of interaction timing and node heterogeneity. In response to this problem, this paper proposed a recommendation method based on dynamic bipartite network representation learning. Firstly, the method constructed a time-series weighted bipartite network, and then respectively mapped user nodes and item nodes to different vector spaces to preserve the heterogeneity of the network, and aggregated the first-order and high-order neighbor information for center nodes with graph convolution. Finally it used a multi-layer perceptron to learn the nonlinear relationship between the two types of node embeddings and performed top-N recommendation. Experimental results on Amazon and Taobao datasets show that the proposed method is significantly superior than the related methods based on static or heterogeneous network representation learning in HR and NDCG indicators.