基于节点重要性分配的网络分类算法
首发时间:2020-02-03
摘要:网络表示学习是将网络节点的内容特征和结构特征结合起来,映射到低维连续向量上的一种学习方法。近年来,端到端图卷积神经网络GCNs被广泛应用于社区检测、链接预测、信息检索、推荐系统等领域。节点通过从邻域节点聚集信息来增强嵌入效果。然而,大多数网络是稀疏的,服从马太效应(幂律分布),高度节点在网络中对于具体任务的效果并不明显。为了提高对于网络节点重要性的学习,本文提出了一种基于GCNs模型的节点重要性评分机制。与传统的模型相比,该模型提升有用节点信息并抑制对当前任务用处不大的节点信息。实验表明,该模型能够提升网络的分类效果。
For information in English, please click here
Network classification algorithm based on node importance distribution
Abstract:Network representation learning is a learning method that combines content features and structure features of network nodes and maps them to low-dimensional continuous vectors. In recent years, end-to-end graph convolutional neural network GCNs has been widely used in community detection, link prediction, information retrieval, recommendation system and other fields. Nodes enhance embedding by aggregating information from neighborhood nodes. However, most networks are sparse and subject to the Matthew effect (power-law distribution), and the effect of height nodes on specific tasks in the network is not obvious. In order to improve the learning of node importance, this paper proposes a GCNs model-based node importance scoring mechanism. Compared with the traditional model, this model promotes useful node information and suppresses node information that is not useful for the current task. Experiments show that this model can improve the classification effect of network.
Keywords: Network representation Sparsity Graph convolutional neural network Node classification
引用
No.****
动态公开评议
共计0人参与
勘误表
基于节点重要性分配的网络分类算法
评论
全部评论0/1000