直推式网络表示学习

发布时间：2018-05-20 22:02

本文选题：直推式 + 网络表示学习　；参考：《计算机科学与探索》2017年04期

【摘要】：网络表示学习是一个经典的学习问题,其目的是将高维的网络在低维度的向量空间进行表示。目前大多数的网络表示学习方法都是无监督的,忽视了标签信息。受LINE(large-scale information network embed-ding)算法启发而提出了一种半监督的学习算法TLINE。TLINE是一种直推式表示学习算法,其通过优化LINE部分的目标函数来保留网络的局部特性。而标签信息部分,则使用线性支持向量机(support vector machine)来提高带标签结点的区分度。通过边采样、负采样和异步随机梯度下降来降低算法的复杂度,从而使TLINE算法可以处理大型的网络。最后,在论文引用数据集Cite Seer和共同作者数据集DBLP上进行了实验,实验结果表明,TLINE算法明显优于经典的无监督网络表示学习算法Deep Walk和LINE。
[Abstract]:Network representation learning is a classical learning problem, which aims to represent high-dimensional networks in low-dimensional vector space. At present, most of the learning methods are unsupervised and ignore the tag information. Inspired by the LINE(large-scale information network embed-ding algorithm, a semi-supervised learning algorithm, TLINE.TLINE, is proposed, which is a direct representation learning algorithm, which preserves the local characteristics of the network by optimizing the objective function of the LINE part. In the part of label information, linear support vector machine is used to improve the discrimination of tagged nodes. The complexity of the algorithm is reduced by edge sampling, negative sampling and asynchronous random gradient descent, so that the TLINE algorithm can handle large networks. Finally, the experiments are carried out on the reference dataset Cite Seer and co-author dataset DBLP. The experimental results show that the proposed algorithm is superior to the classical unsupervised network representation learning algorithms, Deep Walk and LINE, and the experimental results show that the proposed algorithm is superior to the classical unsupervised network representation learning algorithms (Deep Walk and LINE).
【作者单位】：北京大学信息科学技术学院;
【基金】：国家自然科学基金Nos.61272340,U1536201 国家重点基础研究发展计划(973计划)No.2014CB340400~~
【分类号】：TP181

【相似文献】