基于随机游走和GAN的异质信息网络表征方法

Heterogeneous Network Representation Learning Based on Random Walk and GAN

近年来,异质信息网络的表征学习逐渐成为了研究热点。已有的研究中,有使用生成对抗网络来应对这个任务的方法,取得了不错的效果。但是该方法未能有效地利用节点的上下文语义信息。为此,论文提出一种结合生成对抗网络和截断随机游走思想的异质信息网络表征学习方法。首先使用截断随机游走方法获取一个节点的上下文信息,也即游走出的路径,然后基于这些路径借助生成对抗网络来训练模型。基于标准数据集的实验也证明了论文方法的有效性。

In recent years,representational learning of heterogeneous information networks has gradually become a research hotspot.In the previous studies,there is a method to deal with it by using generative adversarial network,which has achieved good results.However,this method fails to make use of the node's contextual semantic information effectively.Therefore,this paper proposes an approach combining generative adversarial network and truncated random walk to apply to representation learning of heterogeneous information network.Firstly,the context of a node is obtained through truncated random walk algorithm,it is the traversed paths.Then based on these paths,the model is trained under the framework of generative adversarial network.Experiments based on standard data sets show the effectiveness of our approach.