【摘要】 网络链接预测是指通过网络结构信息及节点属性信息等网络历史信息预测2个节点之间产生新的链接关系的可能性。网络链接预测是网络分析的基础任务,在异常检测、推荐系统等方面有重要应用。网络表示学习旨在通过无监督方法,将符号化的数据编码到低维、稠密的向量空间中,从而更好地应用于机器学习任务中。由于真实网络数据极其稀疏,现有的模型在链接预测的表现上存在一定的提升空间。针对该问题,文章提出一种基于网络表示学习的属性网络链接预测算法(attributed network embedding based link prediction,ANE-LP)。首先有效提取网络结构信息和节点属性信息,并且通过深度网络结构将网络中各节点表征到低维、稠密向量空间;然后通过相似度度量模型重新定义出邻居节点间的关系;最后在2个真实数据集上进行实验验证。实验结果表明,基于网络特征学习的链接预测算法与其他方法相比更优越。更多还原

【Abstract】 Link prediction is the task of predicting the possible link relationship between nodes based on the network structure information and node attribute information. Link prediction is the basis of network analysis tasks, and has been applied in the anomaly detection, recommendation and so on. Network representation learning aims to encode symbolized data into low-dimensional, dense vector spaces through unsupervised methods. Nevertheless, due to the complexity and sparsity of the network data, the performances of link prediction models are not satisfactory. To this end, a new model named ANE-LP(attributed network embedding based link prediction) is proposed. Specifically, to deal with the problem of data sparsity, ANE-LP takes structure and attribute information into consideration and designs a deep neural network to exploit the non-linear complex information of each node for network embedding. Then, the similarity measurement function is used to calculate the predicted relationship. The experiments on two real datasets demonstrate that the proposed method outperforms the state-of-the-art methods.更多还原