基于Listwise的新型排序算法.pdfVIP

第37 卷 第23 期 计 算 机 工 程 2011 年12 月 Vol.37 No.23 Computer Engineering December 2011 ·人工智能及识别技术· 文章编号：1000—3428(2011)23—0165—03 文献标识码：A 中图分类号：TP181 基于Listwise 的新型排序算法 程 凡，李龙澍 (安徽大学计算机科学与技术学院，合肥 230039) 摘 要：基于Pairwise 的排序算法得到的判别式模型准确率较低。为此，提出一种基于Listwise 的新型排序算法。采用判别式模型，将基 于 1-slack 的支持向量机作为算法框架，定义算法的优化目标。由于该目标的约束条件太多，难以直接优化，因此使用割平面法求解。对 于算法内部寻找最违背排列的子问题，将其看作一个线性指派问题，采用匈牙利法求解。在基准数据集上的实验结果验证该算法的有效性 和稳定性。 关键词：排序算法；结构化学习；Listwise 法；支持向量机；匈牙利法 Novel Ranking Algorithm Based on Listwise CHENG Fan, LI Long-shu (School of Computer Science and Technology, Anhui University, Hefei 230039, China) 【Abstract 】The discriminant model learned by traditional ranking algorithm based on Pairwise method is not accurate. Aiming at the above problem, a novel ranking algorithm based on Listwise method is proposed. The algorithm uses the discriminative model and adopts one-slack Support Vector Machine(SVM) as its framework. On this basis, the algorithm defines its objective function. In order to solve the problem of exponential size of constraints in the optimization, a cutting plane algorithm is introduced. For the sub-problem of finding the most violated constraints, the paper presents to view it as a linear assignment problem, and adopts Hungarian method to solve it. Experimental results on the benchmar