Cost-Sensitive ListMLE Ranking Approach Based on Sparse Representation

摘要

Learning-to-rank plays a pivotal role in information retrieval. To emphasize the top training of the permutation and improve the accuracy of the ranking model, several cost-sensitive listwise ranking algorithms have been proposed by incorporating the cost-sensitive learning idea into the ranking model. However, these methods ignore the impact of the high-dimensional features of the sample on the complexity of model, which results in low computational efficiency of the model. In this article, we proposed a cost-sensitive ListMLE ranking algorithm based on sparse representation which takes into account both the accuracy and computational efficiency of the ranking model. For the sake of achieving sparsity, the l(1) regularized sparse term is added to the existing cost-sensitive ListMLE ranking model, and the global optimal parameters of the model are obtained by a simple yet efficient proximal gradient descent (PGD) learning method. Experiments performed on several benchmark datasets demonstrate that the proposed algorithm can improve empirical performance accuracy in building sparse model.