基于边际的信息检索排序算法研究

Research on margin-based IR ranking algorithm

系统地分析RLR算法模型的优缺点,证明阶铰链亏损函数是凸函数,且满足2/2≥,变换模型框架中的亏损函数,采用减少上下界差距的策略选取参数,由证明阶铰链亏损函数满足+<2+进而得到新的算法。实验结果表明,该算法是有效的,最后讨论可继续研究的课题。

The advantages and disadvantages of RLR algorithm framework are systematically analyzed. It is proved thatp-order hinge loss fimction is convex and satisfies 2L（x/2） ≥ L（x）. A new algorithm is obtained by shifting loss function, choosing a to minimize the gaps between the lower bound and the upper bound and proving thatp-order hinge loss function satisfies L（x）＋L（-x）〈（2＋｜x｜）^p. Experiments show that the new algorithm is effective and future research to be carried out is also discussed.