采用组合方法进行链路预测的理论极限研究

Theoretical limit of link prediction using a combination method

对链路预测组合方法是否存在理论极限以及如何抵近极限开展研究。从是否使用多维度信息或是否直接定义多维度信息之间关系的角度,将链路预测方法分为单机制方法和组合方法。采用简单函数列逼近可测函数的方法,得出链路预测组合方法的理论极限定理;提出使组合方法准确性达到理论上限的组合规则,并给出所提组合规则的几何解释和针对极限定理的仿真示例说明。极限定理揭示了组合方法的本质和组合方法相比单机制方法具有更高准确性及稳健性的原因。

The problem that whether there a theoretical limit exists for link prediction combination methods and how to approximate was investigated.Link prediction methods were divided into single or combination methods,based on whether multidimension information was used,or whether the relation of multidimension information was defined directly.Limit theorems for link prediction by approximating a measurable function by a simple function sequence were provided.Combination rule and corresponding geometric interpretations and simulation examples for limit theorems were also provided.Limit theorems show why combination methods have higher accuracy and robustness than single methods.