基于海量数据的链接预测方法研究

海量数据带来的高复杂性已经成为社会网络分析中不可回避的问题,在链接预测领域,寻找新的高效快速相似度特征算法成为了解决上述问题的关键。文章提出邻居相似度度量,实验证明提高了预测准确率,并保持了较低的计算复杂度。同时针对传统1:1抽样训练样本的方法提出改进,提出1:3抽样。实验结果表明1:3抽样方法有效改善了1:1抽样中存在的欠训练问题。

Research on the Link Prediction Methods Based on Massive Data

The high complexity brought from massive data has become an unavoidable problem in the social network analysis. In link prediction, finding new fast and efficient algorithms for similarity characteristics becomes the key to solve the above problem. This paper proposes a neighbor affinity measure, experimentally proves the improvement in the prediction accuracy, and a lower computational complexity. Also the paper improves the method of traditional 1：1 sampling strategy, and proposes the 1：3 sampling strategy. The experiment shows that the new sampling method effectively ameliorates the problem existing in the 1：1 sampling.