一种基于孪生神经网络的深度异构网络嵌入算法

针对现有异构网络嵌入方法导致的捕获关系冗余和模糊的问题，提出了一种基于孪生神经网络的深度异构网络嵌入模型。首先，基于面向关系的深度嵌入(Relation-Oriented Deep Embedding，RODE)框架构建了异构网络嵌入模型，以区分同型节点和异型节点之间的关系；其次，将同型节点与异类节点之间的相似性近似到低维空间，通过构建多任务的孪生神经网络来实现节点之间结构和语义关系的深度嵌入；最后，选取四个数据集执行典型网络挖掘任务，并与其他六种算法进行实验对比分析。实验结果表明，保持相同类型节点之间的相似性有助于提高节点分类效率，且损失函数在提高异构网络嵌入质量方面具有良好的优越性；RODE模型能够有效提高稀疏网络的嵌入质量，且具有良好的稳定性和鲁棒性。

A Deep Heterogeneous Network Embedding Algorithm Based on Twin Neural Network

For the problem of redundancy and ambiguity of capturing relationship caused by existing heterogeneous network embedding methods,a deep heterogeneous network embedding based on twin neural network is proposed.First,a heterogeneous network embedding model is constructed based on the relation-oriented deep embedding(RODE) framework.The relationship between homogeneous nodes and heterogeneous nodes is distinguished by a heterogeneous network model.Second,the similarity between nodes of the same type and nodes of different types is approximated to a low-dimensional space.The deep embedding of the structure and semantic relationship between nodes is realized by constructing a multi-task twin neural network.Finally,four data sets are selected to perform typical network mining tasks,and compared with other six algorithms.Experimental results show that the similarity between nodes of the same type is maintained to improve the efficiency of node classification,and the loss function has good advantages in improving the quality of heterogeneous network embedding.RODE model can effectively improve the embedding quality of sparse networks,and has good stability and robustness.