融合朴素贝叶斯方法的复杂网络链路预测

近来复杂网络成为了众多学者的研究热点。但真实网络中的连边信息并不完整，不利于网络的分析研究，链路预测可以挖掘网络中的缺失连边，为网络重构提供基本依据。本文认为网络中链接的产生不仅受外部因素――共同邻居的影响，还受其自身因素的影响。其中，共同邻居的影响可以通过文献中的局部朴素贝叶斯（LNB）模型量化，节点的影响则根据其自身的度量化。本文将两者综合考虑，提出了融合朴素贝叶斯（SNB）模型，然后用共同邻居（CN）、Adamic-Adar（AA）和资源分配（RA）指标进行推广。在美国航空网（USAir）上的实验结果表明，该方法的预测准确度比LNB和基准方法均有所提高，从而证明了该方法的有效性。
Recently, complex networks have become a research hotspot. However, edge information in the real network is incomplete, which is not conducive to the analysis and research of the network. Link prediction can provide a fundamental basis for network reconstruction by digging out the missing edges in the network. This paper demonstrates that the generation of links in the network is not only influenced by external factors (common neighbors) but also by its own factors. Among them, the influence of common neighbors can be quantified via the local naive Bayes (LNB) model in the literature, whereas the influence of nodes can be quantified depending on their degree. Therefore, a syncretic naive Bayes (SNB) model is proposed based on comprehensive consideration of the influence of the two abovementioned aspects. The model is then extended to common neighbors, Adamic-Adar, and Resource Allocation methods. Finally, the experimental results on USAir show that the prediction accuracy of the method is higher than that of LNB and the benchmark method, which proves the effectiveness of the SNB model