研究了在样本集中删除一个元素后二部排序算法的稳定性。证明了如果使用lγ作为排序亏损函数,那么算法在具有得分一致稳定的前提下也具有亏损一致稳定;并且对任意x,K(x,x)存在有限上界。通过最小化正则经验l-误差得到的排序算法具有良好的一致得分稳定。
The stability of bipartite ranking algorithms is studied when an element is deleted from the sample set altogether.If the ranking algorithm is used as a loss function,it has both the uniform loss stability and uniform score stability;and it has a limited upper bound for any x,K(x,x).The ranking algorithm by minimizing the regularization empirical l-error has good uniform score stability.