Learning to Rank with Nonsmooth Cost Functions
Learning to Rank with Nonsmooth Cost Functions
复制标题
DOI：
10.7551/mitpress/7503.003.0029
复制
发表时间：
2006-12
期刊：
影响因子：
--
通讯作者：
C. Burges;R. Ragno;Quoc V. Le
中科院分区：
文献类型：
--
作者：
C. Burges;R. Ragno;Quoc V. Le
信息检索中使用的质量度量特别难以直接优化，因为它们仅通过给定查询返回的文档的排序顺序依赖于模型分数。因此，成本相对于模型参数的导数要么为零，要么未定义。在本文中，我们提出了一类简单，灵活的算法，称为LambdaRank，它避免了这些困难的隐式成本函数。我们使用神经网络模型描述LambdaRank，尽管这个想法适用于任何可微函数类。我们得到的隐式成本函数是凸的充分必要条件，我们表明，一般的方法有一个简单的机械解释。我们证明了显着提高的准确性，在一个国家的最先进的排名算法，在几个数据集。我们还表明，LambdaRank提供了一种方法，可以显着加快该排名算法的训练阶段。虽然本文是针对排名，所提出的方法可以扩展到任何非光滑和多元成本函数。
The quality measures used in information retrieval are particularly difficult to optimize directly, since they depend on the model scores only through the sorted order of the documents returned for a given query. Thus, the derivatives of the cost with respect to the model parameters are either zero, or are undefined. In this paper, we propose a class of simple, flexible algorithms, called LambdaRank, which avoids these difficulties by working with implicit cost functions. We describe LambdaRank using neural network models, although the idea applies to any differentiable function class. We give necessary and sufficient conditions for the resulting implicit cost function to be convex, and we show that the general method has a simple mechanical interpretation. We demonstrate significantly improved accuracy, over a state-of-the-art ranking algorithm, on several datasets. We also show that LambdaRank provides a method for significantly speeding up the training phase of that ranking algorithm. Although this paper is directed towards ranking, the proposed method can be extended to any non-smooth and multivariate cost functions.