Abstract: As one of the important research directions of complex networks, it is a commonly used method based on node structure similarity to predict. There are lots of weak clique structures in real networks. The key problem of link prediction is to build algorithms for different network structures. Taking advantage of friends recommendation strategies in social networks, introducers tend to introduce their more familiar people to acceptors. This paper proposes a node similarity measure index which is more suitable for a particular type of weak clique structure because it combines local feature description and differentiates the difference between the influences of the user nodes effectively. The experimental results of the proposed friend recommendation model link prediction algorithm according to the index on 12 datasets show that the algorithm has obvious advantages on the two evaluation criteria of AUC and Precision.

Key words: complex networks, friend recommendation, link prediction, similarity index

摘要： 链路预测是复杂网络的一个重要研究方向。基于节点结构相似性进行链路预测是目前常用的方法。真实网络中存在大量的局部群落结构，针对不同的网络结构构建算法是链路预测的核心问题。利用社交网络好友推荐策略，中介人倾向于将自己更熟悉的人介绍给目标用户，提出了一种节点相似性度量指标。该指标结合局部特征描述并有效区分了用户节点之间影响力的不同，更适用于一类特定的局部群落结构。依据该指标提出的加权好友推荐模型链路预测算法在12个数据集上的实验结果表明，该算法在AUC和Precision两个评价标准上具有明显优势。

关键词: 复杂网络, 好友推荐, 链路预测, 相似性指标