基于节点地位和相似性的社交网络边符号预测

边符号预测即根据网络拓扑结构挖掘符号相关隐含信息，旨在揭示用户之间的潜在关系。节点地位和相似性能够较好地体现边符号属性，为改善预测效果提供了理论基础。通过探究二者与边符号属性之间的强相关性，建立符号预测模型。首先，利用排序算法prestige评估用户节点的社会地位，同时使用余弦相似度表示用户的社交偏好；然后，在逻辑回归学习模型的基础上融合二者建立边符号预测模型LR-SN；最后，在模型训练过程中采用随机梯度上升算法优化求解。三个真实网络数据集的实验结果表明，相比于现有基准方法，LR-SN模型的符号预测准确率显著提高且具有一定的推广性，说明通过融合局部信息与全局信息能够进一步改善预测效果。

Social network edge sign prediction based on node status and similarity

The edge sign prediction is to mine the sign-related implicit information according to the network topology, aiming to reveal the potential relationship between users. Node status and similarity can better represent sign attributes of edges, providing a theoretical basis for improving the prediction effect. By investigating the strong correlation between the two theories and the sign attributes of the edges, this paper established a sign prediction model. Firstly, it used prestige evaluate the social status of user nodes. At the same time, cosine similarity could represent the user''s social preferences. Then, both sides were combined based on the logistic regression learning model to establish the edge sign prediction model LR-SN. Finally, a random gradient ascent algorithm would optimize the model during training. The experimental results of three real network datasets show that compared with the existing baseline methods, the accuracy of sign prediction of LR-SN model is significantly improved and has certain generalization, indicating that the fusion of local information and global information can further improve the prediction effect.